Plug some leaks in curl's proxy handling

Fix Issue #4047 Check return codes and free objects

travis: add custom apt sources

remote: add typedef to normalize push_update_reference callback

refs: do not use peeled OID if peeling to a tag

Move back to Travis's VM infrastructure for efficiency.

Very many callbacks in libgit2 have some sort of typedef to normalize the name at git_<name_of_operation>_cb. Add a typedef for push_update_references in the remote so the name follows the same conventions.

Document that a commit is not a descendant of itself

Use SOCK_CLOEXEC when creating sockets

If a reference stored in a packed-refs file does not directly point to a
commit, tree or blob, the packed-refs file will also will include a
fully-peeled OID pointing to the first underlying object of that type.
If we try to peel a reference to an object, we will use that peeled OID
to speed up resolving the object.

As a reference for an annotated tag does not directly point to a commit,
tree or blob but instead to the tag object, the packed-refs file will
have an accomodating fully-peeled OID pointing to the object referenced
by that tag. When we use the fully-peeled OID pointing to the referenced
object when peeling, we obviously cannot peel that to the tag anymore.

Fix this issue by not using the fully-peeled OID whenever we want to
peel to a tag. Note that this does not include the case where we want to
resolve to _any_ object type. Existing code may make use from the fact
that we resolve those to commit objects instead of tag objects, even
though that behaviour is inconsistent between packed and loose
references. Furthermore, some tests of ours make the assumption that we
in fact resolve those references to a commit.

Static linking for bundled deps

The block in the script installs the packages if we're _not_ on Precise. This
was dropped in c17c3f8a ("travis: drop support for Ubuntu Precise") in error.

When we duplicate a user-provided options struct, we're stuck with freeing the
url in it. In case we add stuff to the proxy struct, let's add a function in
which to put the logic.

Our bundled deps are being built as simple static libraries which are
then linked into the libgit2 library via `TARGET_LINK_LIBRARIES`. While
this works for a dynamically built libgit2 library, using this function
to link two static libraries does not have the expected outcome of
merging those static libraries into one big library. This leads to
symbols of our bundled deps being undefined in the resulting libgit2
archive.

As we have bumped our minimum CMake version to 2.8.11, we can now easily
make use of object libraries for our bundled dependencies. So build
instructions are still self-contained inside of the dependency
directories and the resulting object libraries can just be added to the
LIBGIT2_OBJECTS list, which will cause them to be linked into the final
resulting static library. This fixes the issue of undefined symbols.

There are two locations where we check whether CMake supports
`TARGET_INCLUDE_DIRECTORIES`. While the first one uses `VERSION_LESS
2.8.12`, the second one uses `VERSION_GREATER 2.8.11`, which are
obviously equivalent to each other. It'd still be easier to grep for
specific CMake versions being required for some features if both used
the same conditional mentioning the actual target version required. So
this commit refactors these conditions to make them equal.

As we have bumped our minimum CMake version to 2.8.11, we can now
unconditionally make use of object libraries. So remove the version
check for the git2internal object library and always use it.

Our current minimum CMake version is 2.8. This version does not yet
allow us to use object libraries (introduced in 2.8.8) and target
include directories (introduced in 2.8.12), which are both mechanisms we
want to use to fix some specific problems. We previously were not able
to bump our CMake version to a version supporting object libraries
because Ubuntu Precise only had CMake version 2.8.7 in its repositories.
But due to Precise being end of life now, we shouldn't need to honor it
anymore. A current survey of some of the more conservative distributions
brings up the following versions of CMake:

- CentOS 5: 2.6.2
- CentOS 6: 2.8.12.2
- Debian 7: 2.8.11
- Fedora 23: 3.3.2
- OpenSUSE 13.2: 3.0.2
- Ubuntu Precise: 2.8.7
- Ubuntu Trusty: 2.8.12

The only two outliers here are CentOS 5 and Ubuntu Precise. CentOS is
currently unsupported due to our minimum version being 2.8 and Ubuntu
Precise is not maintained anymore. So the next smallest version
supported by all major distributions is 2.8.11. While this does not yet
support target include directories, it at least enables us to use object
libraries. So this becomes our new minimum required version.

Distinguish variables keeping track of our internal libgit2 sources and
the final objects which shall be linked into the library. This will ease
the transition to use object libraries for our bundled dependencies
instead of linking them in.

Ubuntu Precise is end of life since April 2017. At that point in time,
Precise was still the main distro on which Travis CI built upon, with
the Trusty-based images still being in a beta state. But since June
21st, Trusty has officially moved out of beta and is now the default
image for all new builds. Right now, we build on both old and new images
to assure we support both.

Unfortunately, this leaves us with the highest minimum version for CMake
being 2.8.7, as Precise has no greater version in its repositories. And
because of this limitation, we cannot actually use object libraries in
our build instructions. But considering Precise is end of life and
Trusty is now the new default for Travis, we can and should drop support
for this old and unmaintained distribution. And so we do.

Reproducible builds

diff: cleanup hash ctx in `git_diff_patchid`

Fix AppVeyor build failures due to CRTDBG linking issue

In order to cover a wider range of build environments, add two more jobs
which build and test libgit2 on Visual Studio 14 2015.

AppVeyor currently does provide three standard build worker images with
VS2013, VS2015 and VS2017. Right now, we are using the implicitly, which
is the VS2015 one. We want to be more explicit about this, so that we
can easily switch build images based on the job. So starting from this
commit, we explicitly set the `APPVEYOR_BUILD_WORKER_IMAGE` variable per
job, which enables us to choose different images.

To be able to test a wider range of build configurations, this commit
also switches the jobs for VC2010 over to use the older, VS2013 based
images. As the next commit will introduce two new jobs for building with
VS2015, we have then covered both build environments.

Also, let us be a bit more explicit regarding the CMake generator.
Instead of only saying "Visual Studio 10", use the more descriptive
value "Visual Studio 10 2010" to at least avoid some confusion
surrounding the versioning scheme of Visual Studio.

When the MSVC_CRTDBG option is set by the developer, we will link in the
dbghelper library to enable memory lead detection in MSVC projects. We
are doing so by adding it to the variable `CMAKE_C_STANDARD_LIBRARIES`,
so that it is linked for every library and executable built by CMake.
But this causes our builds to fail with a linker error:

```
    LINK: fatal error LNK1104: cannot open file 'advapi32.lib;Dbghelp.lib'
```

The issue here is that we are treating the variable as if it were an
array of libraries by setting it via the following command:

```
    SET(CMAKE_C_STANDARD_LIBRARIES "${CMAKE_C_STANDARD_LIBRARIES}"
        "Dbghelp.lib")
```

The generated build commands will then simply stringify the variable,
concatenating all the contained libraries with a ";". This causes the
observed linking failure.

To fix the issue, we should just treat the variabable as a simple
string. So instead of adding multiple members, we just add the
"Dbghelp.lib" library to the existing string, separated by a space
character.

After initializing the hash context in `git_diff_patchid`, we never
proceed to call `git_hash_ctx_cleanup` on it. While this doesn't really
matter on most hash implementations, this causes a memory leak on Win32
due to CNG system requiring a `malloc` call.

Fix the memory leak by always calling `git_hash_ctx_cleanup` before
exiting.

By default, both ar(1) and ranlib(1) will insert additional information
like timestamps into generated static archives and indices. As a
consequence, generated static archives are not deterministic when
created with default parameters.

Both programs do support a deterministic mode, which will simply zero
out undeterministic information with `ar D` and `ranlib -D`.
Unfortunately, CMake does not provide an easy knob to add these command
line parameters. Instead, we have to redefine the complete command
definitons stored in the variables CMAKE_C_ARCHIVE_CREATE,
CMAKE_C_ARCHIVE_APPEND and CMAKE_C_ARCHIVE_FINISH.

Introduce a new build option `ENABLE_REPRODUCIBLE_BUILDS`. This option
is available on Unix-like systems with the exception of macOS, which
does not have support for the required flags. If the option is being
enabled, we add those flags to the invocation of both `ar` and `ranlib`
to enable deterministically building the static archive.

Clear the remote_ref_name buffer in git_push_update_tips()

features.h: allow building without CMake-generated feature header

In commit a390a846 (cmake: move defines into "features.h" header,
2017-07-01), we have introduced a new "features.h" header. This file is
being generated by the CMake build system based on how the libgit2 build
has been configured, replacing the preexisting method of simply setting
the defines inside of the CMake build system. This was done to help
splitting up the build instructions into multiple separate
subdirectories.

An overlooked shortcoming of this approach is that some projects making
use of libgit2 build the library with custom build systems, without
making use of CMake. For those users, the introduction of the
"features.h" file makes their life harder as they would have to also
generate this file.

Fix this issue by guarding all inclusions of the generated header file
by the `LIBGIT2_NO_FEATURES_H` define. Like this, other build systems
can skip the feature header and simply define all used features by
specifying `-D` flags for the compiler again.

If fetch_spec was a non-pattern, and it is not the first iteration of push_status vector, then git_refspec_transform would result in the new value appended via git_buf_puts to the previous iteration value.

Forcibly clearing the buffer on each iteration to prevent this behavior.

README: Mention Guile-Git bindings.

The script "generate.py" is used to parse all test source files for unit
tests. These are then written into a "clar.suite" file, which can be
included by the main test executable to make available all test suites
and unit tests.

Our current algorithm simply collects all test suites inside of a dict,
iterates through its items and dumps them in a special format into the
file. As the order is not guaranteed to be deterministic for Python
dictionaries, this may result in arbitrarily ordered C structs. This
obviously defeats the purpose of reproducible builds, where the same
input should always result in the exact same output.

Fix this issue by sorting the test suites by name previous to dumping
them as structs. This enables reproducible builds for the libgit2_clar
file.