The windows sysdir functions are libgit2-specific and for git
compatibility only; remove them from the shared util directory and bring
them into the libgit2 source tree.

Instead of simply including the utility files directly, make them a
cmake object library for easy reusability between other projects within
libgit2.

Now the top-level `src` is responsible for platform selection, while the
next-level `libgit2` and `util` configurations are responsible for
identifying what objects they include.

`mktemp` on mingw is exceedingly deficient, using a single monotonically
increasing alphabetic character and the pid.  We need to use our own
random number generator for temporary filenames.

Previously, we were using `umask(mask = umask(0))` to fetch the current
umask in order to apply it to the desired mode, but this is broken in
the presence of threads. There is no portable way to directly fetch
umask without mutating it.

Instead, add a reimplementation of mkstemp that uses a passed-in mode,
instead of explicitly chmodding to 0600 like POSIX requires of mkstemp.

Fixes: jmgao/pore#46

Remove some unnecessary includes from utility code.

Introduce `git_fs_path`, which operates on generic filesystem paths.
`git_path` will be kept for only git-specific path functionality (for
example, checking for `.git` in a path).

libgit2 has two distinct requirements that were previously solved by
`git_buf`.  We require:

1. A general purpose string class that provides a number of utility APIs
   for manipulating data (eg, concatenating, truncating, etc).
2. A structure that we can use to return strings to callers that they
   can take ownership of.

By using a single class (`git_buf`) for both of these purposes, we have
confused the API to the point that refactorings are difficult and
reasoning about correctness is also difficult.

Move the utility class `git_buf` to be called `git_str`: this represents
its general purpose, as an internal string buffer class.  The name also
is an homage to Junio Hamano ("gitstr").

The public API remains `git_buf`, and has a much smaller footprint.  It
is generally only used as an "out" param with strict requirements that
follow the documentation.  (Exceptions exist for some legacy APIs to
avoid breaking callers unnecessarily.)

Utility functions exist to convert a user-specified `git_buf` to a
`git_str` so that we can call internal functions, then converting it
back again.

In `git_futils_readbuffer_updated`, always take a particular hash
instead of a `git_oid`.  This lets us change the checksum algorithm
independently of `git_oid` usage.

Separate the concerns of the hash functions from the git_oid functions.
The git_oid structure will need to understand either SHA1 or SHA256; the
hash functions should only deal with the appropriate one of these.

Provide a mechanism for system components to register for initialization
and shutdown of the libgit2 runtime.

Move the mwindow mutex into the mwindow code itself, initializing it in
the mwindow global initialization function instead of in the global
initializer.

While the function `git_futils_fake_symlink` is declared with arguments
`new, old`, the implementation uses the reverse order `old, new`. Let's
fix the ordering issues to be `new, old` for both, which matches what
symlink(3P) has. While at it, we also rename these parameters: `old` and
`new` doesn't really make a lot of sense in the context of symlinks,
which is why this commit renames them to be called `target` and `path`.

Prefer `off64_t` to `git_off_t` for internal visibility.

Instead of using a signed type (`off_t`) use an unsigned `uint64_t` for
the size of the files.

Instead of using a signed type (`off_t`) use `uint64_t` for the maximum
size of files.

IIRC I got a strange return once from lstat, which translated in a weird
error class/message being reported. As a safety measure, enforce a -1 return in
that case.

Our file utils functions all have a "futils" prefix, e.g.
`git_futils_touch`. One would thus naturally guess that their
definitions and implementation would live in files "futils.h" and
"futils.c", respectively, but in fact they live in "fileops.h".

Rename the files to match expectations.

In commit 45f24e78 (git_repository_init: stop traversing at
windows root, 2019-04-12), we have fixed `git_futils_mkdir` to
correctly handle the case where we create a directory in
Windows-style filesystem roots like "C:\repo".

The problem here is an off-by-one: previously, to that commit,
we've been checking wether the parent directory's length is equal
to the root directory's length incremented by one. When we call
the function with "/example", then the parent directory's length
("/") is 1, but the root directory offset is 0 as the path is
directly rooted without a drive prefix. This resulted in `1 == 0 +
1`, which was true. With the change, we've stopped incrementing
the root directory length, and thus now compare `1 <= 0`, which
is false.

The previous way of doing it was kind of finicky any non-obvious,
which is also why the error was introduced. So instead of just
re-adding the increment, let's explicitly add a condition that
aborts finding the parent if the current parent path is "/".

Making this change causes Azure Pipelines to fail the testcase
repo::init::nonexistent_paths on Unix-based systems. This is because we
have just fixed creating directories in the filesystem root, which
previously didn't work. As Docker-based tests are running as root user,
we are thus able to create the non-existing path and will now succeed to
create the repository that was expected to actually fail.

Let's split this up into three different tests:

- A test to verify that we do not create repos in a non-existing parent
  directoy if the flag `GIT_REPOSITORY_INIT_MKPATH` is not set.

- A test to verify that we fail if the root directory does not exist. As
  there is a common root directory on Unix-based systems that always
  exist, we can only test for this on Windows-based systems.

- A test to verify that we fail if trying to create a repository in an
  unwriteable parent directory. We can only test this if not running
  tests as root user, as CAP_DAC_OVERRIDE will cause us to ignore
  permissions when creating files.

Stop traversing the filesystem at the Windows directory root.  We were
calculating the filesystem root for the given directory to create, and
walking up the filesystem hierarchy.  We intended to stop when the
traversal path length is equal to the root path length (ie, stopping at
the root, since no path may be shorter than the root path).

However, on Windows, the root path may be specified in two different
ways, as either `Z:` or `Z:\`, where `Z:` is the current drive letter.
`git_path_dirname_r` returns the path _without_ a trailing slash, even
for the Windows root.  As a result, during traversal, we need to test
that the traversal path is _less than or equal to_ the root path length
to determine if we've hit the root to ensure that we stop when our
traversal path is `Z:` and our calculated root path was `Z:\`.

Currently, one would use the function `git_strmap_insert` to insert key/value
pairs into a map. This function has historically been a macro, which is why its
syntax is kind of weird: instead of returning an error code directly, it instead
has to be passed a pointer to where the return value shall be stored. This does
not match libgit2's common idiom of directly returning error codes.

Introduce a new function `git_strmap_set`, which takes as parameters the map,
key and value and directly returns an error code. Convert all callers of
`git_strmap_insert` to make use of it.

Move to the `git_error` name in the internal API for error-related
functions.

We want to do this in order to get FETCH_HEAD to be empty when we start updating
it due to fetching from the remote.

Next to including several files, our "common.h" header also declares
various macros which are then used throughout the project. As such, we
have to make sure to always include this file first in all
implementation files. Otherwise, we might encounter problems or even
silent behavioural differences due to macros or defines not being
defined as they should be. So in fact, our header and implementation
files should make sure to always include "common.h" first.

This commit does so by establishing a common include pattern. Header
files inside of "src" will now always include "common.h" as its first
other file, separated by a newline from all the other includes to make
it stand out as special. There are two cases for the implementation
files. If they do have a matching header file, they will always include
this one first, leading to "common.h" being transitively included as
first file. If they do not have a matching header file, they instead
include "common.h" as first file themselves.

This fixes the outlined problems and will become our standard practice
for header and source files inside of the "src/" from now on.

Only ignore `EBUSY` from `rmdir` when the `GIT_RMDIR_SKIP_NONEMPTY` bit
is set.

When the `git_futils_mmap_ro_file` function encounters an error after
the file has been opened, it will do a simple returns. Instead, we
should close the opened file descriptor to avoid a leak. This commit
fixes the issue.

When executing `git_futils_mmap_ro_file`, we first try to guess whether
the file is mmapable at all. Part of this check is whether the file is
too large to be mmaped, which can be true on systems with 32 bit
`size_t` types.

The check is performed by first getting the file size wtih
`git_futils_filesize` and then checking whether the returned size can be
represented as `size_t`, returning an error if so. While this test also
catches the case where the function returned an error (as `-1` is not
representable by `size_t`), we will set the misleading error message
"file too large to mmap". But in fact, a negative return value from
`git_futils_filesize` will be caused by the inability to fstat the file.

Fix the error message by handling negative return values separately and
not overwriting the error message in that case.

`O_EXCL` and `O_TRUNC` are mutually exclusive flags to open(2); you can't
truncate a file if you're asserting that it can't exist in the first place.
Drop `O_TRUNC`.

Don't compute the sha-1 in `git_futils_readbuffer_updated` unless the
checksum was requested.  This means that `git_futils_readbuffer` will
not calculate the checksum unnecessarily.

Windows doesn't support it.

When fsync'ing files, fsync the parent directory in the case where we
rename a file into place, or create a new file, to ensure that the
directory entry is flushed correctly.

Only use defaults for `git_futils_writebuffer` when flags == 0, lest
(1 << 31) be treated as the defaults.

Add a custom `O_FSYNC` bit (if it's not been defined by the operating
system`) so that `git_futils_writebuffer` can optionally do an `fsync`
when it's done writing.

We call `fsync` ourselves, even on systems that define `O_FSYNC` because
its definition is no guarantee of its actual support.  Mac, for
instance, defines it but doesn't support it in an `open(2)` call.