Regression introduced in commit 5452e49f on PR #4967.

Signed-off-by: Sven Strickroth <email@cs-ware.de>

hash: fix missing error return on production builds

When no hash algorithm has been initialized in a given hash context,
then we will simply `assert` and not return a value at all. This works
just fine in debug builds, but on non-debug builds the assert will be
converted to a no-op and thus we do not have a proper return value.

Fix this by returning an error code in addition to the asserts.

Resolve static check warnings in example code

Using cppcheck on libgit2 sources indicated two warnings in
example code.

merge.c was reported as having a memory leak. Fix applied
was to `free()` memory pointed to by `parents`.

init.c was reported as having a null pointer dereference
on variable arg. Function 'usage' was being called with
a null variable. Changed supplied parameter to empty string.

Multiple hash algorithms

More documentation

Incomplete commondir support

For example, https://git-scm.com/docs/gitrepository-layout says:

    info
    Additional information about the repository is recorded in this directory.
    This directory is ignored if $GIT_COMMON_DIR is set
    and "$GIT_COMMON_DIR/info" will be used instead.

So when looking for `info/attributes`, we need to check the commondir first,
or fallback to "our" `info/attributes`.

As with the preceding commit, the ignore code tries to load code from
info/exclude, and we fail to ignore a non-existent file here.

If creating a repository without a common directory (e.g. by
using `git_repository_new`), then `git_repository_item_path` will
return `GIT_ENOTFOUND` for every file that's usually located in
this directory. While we do not care for this case when looking
up the "info/attributes" file, we fail to properly ignore these
errors when setting up or collecting attributes files. Thus, the
gitattributes lookup is broken and will only ever return
`GIT_ENOTFOUND`.

Fix this issue by properly ignoring `GIT_ENOTFOUND` returned by
`git_repository_item_path`.

The code in the `attr_setup` function is not really matching our
current coding style. Besides alignment issues, it's also hard to
see what functions calls depend on one another because they're
split up over multiple conditional statements.

Fix these issues by grouping together dependent function calls
and adjusting the alignment.

Remove warnings

Re-run flaky tests

Our online tests are occasionally flaky since they hit real network
endpoints.  Re-run them up to 5 times if they fail, to allow us to
avoid having to fail the whole build.

Our online tests are occasionally flaky since they hit real network
endpoints.  Re-run them up to 5 times if they fail, to allow us to
avoid having to fail the whole build.

Create an enum that allows us to distinguish between different
hashing algorithms. This enum is embedded into each
`git_hash_ctx` and will instruct the code to which hashing
function the particular request shall be dispatched.

As we do not yet have multiple hashing algorithms, we simply
initialize the hash algorithm to always be SHA1. At a later
point, we will have to extend the `git_hash_init_ctx` function to
get as parameter which algorithm shall be used.

Create a separate `git_hash_sha1_ctx` structure that is specific
to the SHA1 implementation and move all SHA1 functions over to
use that one instead of the generic `git_hash_ctx`. The
`git_hash_ctx` for now simply has a union containing this single
SHA1 implementation, only, without any mechanism to distinguish
between different algortihms.

As a preparatory step to allow multiple hashing APIs to exist at
the same time, split the hashing functions into one layer for generic
hashing and one layer for SHA1-specific hashing. Right now, this is
simply an additional indirection layer that doesn't yet serve any
purpose. In the future, the generic API will be extended to allow for
choosing which hash to use, though, by simply passing an enum to the
hash context initialization function. This is necessary as a first step
to be ready for Git's move to SHA256.

As we will include additional hash algorithms in the future due
to upstream git discussing a move away from SHA1, we should
accomodate for that and prepare for the move. As a first step,
move all SHA1 implementations into a common subdirectory.

Also, create a SHA1-specific header file that lives inside the
hash folder. This header will contain the SHA1-specific header
includes, function declarations and the SHA1 context structure.

Add the `-Wno-documentation-deprecated-sync` switch when compiling with
clang, since our documentation adds `deprecated` markers, but we do not
add the deprecation attribute in the code itself.  (ie, the code is out
of sync with the docs).

In fact, we do not _want_ to mark these items as deprecated in the code,
at least not yet, as we are not quite ready to bother our end-users with
this since they're not going away.

GetProcAddress is prototyped to return a `FARPROC`, which is meant to be
a generic function pointer.  It's literally `int (FAR WINAPI * FARPROC)()`
which gcc complains if you attempt to cast to a `void (*)(GIT_SRWLOCK *)`.
Cast to a `void *` before casting to avoid warnings about the arguments.

GetProcAddress is prototyped to return a `FARPROC`, which is meant to be
a generic function pointer.  It's literally `int (FAR WINAPI * FARPROC)()`
which gcc complains if you attempt to cast to a `void (*)(GIT_SRWLOCK *)`.
Cast to a `void *` before casting to avoid warnings about the arguments.

MinGW does not define DWORD_MAX.  Specify it when it's not defined.

MinGW uses gcc, which expects POSIX formatting for printf, but uses the
Windows C library, which uses its own format specifiers.  Therefore, it
gets confused about format specifiers.  Disable warnings for format
specifiers.

RtlCaptureStackBackTrace is well-defined in Windows, no need to redefine
it.

For MSVC, support warnings as errors by providing the /WX compiler
flags.  (/WX is the moral equivalent of -Werror.)

Disable warnings as errors ass part of xdiff, since it contains
warnings.  But as a component of git itself, we want to avoid skew and
keep our implementation as similar as possible to theirs.  We'll work
with upstream to fix these issues, but in the meantime, simply let those
continue to warn.

Move `git_win32__file_attribute_to_stat` to a regular function instead
of an inlined function.  This helps avoid header ordering issues and
declarations.

The hash source files have circular include dependencies right
now, which shows by our broken generic hash implementation. The
"hash.h" header declares two functions and the `git_hash_ctx`
typedef before actually including the hash backend header and can
only declare the remaining hash functions after the include due
to possibly static function declarations inside of the
implementation includes.

Let's break this cycle and help maintainability by creating a
real implementation file for each of the hash implementations.
Instead of relying on the exact include order, we now especially
avoid the use of `GIT_INLINE` for function declarations.

The structure `git_hash_prov` is only ever used by the Win32 SHA1
backend. As such, it doesn't make much sense to expose it via the
generic "hash.h" header, as it is an implementation detail of the Win32
backend only. Move the typedef of `git_hash_prov` into
"hash/sha1/win32.h" to fix this.

Ensure that the server has not sent us overly-large sideband messages
(ensure that they are no more than `INT_MAX` bytes), then cast to `int`.