This change does a medium-size refactor of the git_commit_graph_file and
the interaction with the ODB. Now instead of the ODB owning a direct
reference to the git_commit_graph_file, there will be an intermediate
git_commit_graph. The main advantage of that is that now end users can
explicitly set a git_commit_graph that is eagerly checked for errors,
while still being able to lazily use the commit-graph in a regular ODB,
if the file is present.

This change makes revwalks a bit faster by using the `commit-graph` file
(if present). This is thanks to the `commit-graph` allow much faster
parsing of the commit information by requiring near-zero I/O (aside from
reading a few dozen bytes off of a `mmap(2)`-ed file) for each commit,
instead of having to read the ODB, inflate the commit, and parse it.

This is done by modifying `git_commit_list_parse()` and letting it use
the ODB-owned commit-graph file.

Part of: #5757

tree: deprecate `git_treebuilder_write_with_buffer`

winhttp: skip certificate check if unable to send request

[skip ci]

commit-graph: Introduce `git_commit_graph_needs_refresh()`

commit-graph: Support lookups of entries in a commit-graph

The function `git_treebuilder_write_with_buffer` is unnecessary; it
is used internally as part of treebuilder writing, but it has little
use to external callers.  For callers that repeatedly write a
treebuilder, we can supply them with a buffer in the treebuilder struct
instead of recreating it.  For ourselves, when we want a single buffer
in our write loop, we can use an internal function.

merge: Check insert_head_ids error in create_virtual_base

Check git_signature_dup failure

insert_head_ids can fail due to allocation error

git_signature_dup can have an allocation failure

In some circumstances (e.g. when proxies are involved), winhttp will fail to reach the WINHTTP_CALLBACK_STATUS_SENDING_REQUEST phase. If this occurs, we'll error with ERROR_WINHTTP_INCORRECT_HANDLE_STATE when attempting to query the server certificate context (see https://docs.microsoft.com/en-us/windows/win32/api/winhttp/nf-winhttp-winhttpsendrequest#remarks).

To avoid this, verify that WinHttpSendRequest has reached the WINHTTP_CALLBACK_STATUS_SENDING_REQUEST phase before checking the certificate. Since we're using WinHTTP in synchronous mode, we know for sure that once WinHttpSendRequest returns we've either sent it successfully or not.

NOTE: WINHTTP_CALLBACK_STATUS_SENDING_REQUEST appears to be deprecated with no direct replacement. WINHTTP_CALLBACK_STATUS_SENDREQUEST_COMPLETE is only available in async mode, and there doesn't appear to be a method of querying this flag outside of the status callback.

Fix some typos

include: fix typos in comments

Fix documentation formating on repository.h

The enum 'git_repository_init_flag_t', 'git_repository_init_mode_t' and
the structure 'git_repository_init_options' does not follow the format
used by docurium.

index: Check git_vector_dup error in write_entries

This is for extra safety within write_entries

If allocating case_sorted.contents fails, git_vector_sort will segfault.

To prevent a race condition, these wlock/rlock calls should probably be checked
for errors.

Invert the workflow conditionals so that a workflow_dispatch event acts
like an on: push build.

Add new bindings for the R language

Update .gitignore

patch: add owner accessor

commit-graph: Introduce a parser for commit-graph files

revspec: rename git_revparse_mode_t to git_revspec_t

mwindow: Fix a bug in the LRU window finding code

The information about the type of a revision spec is not information
about the parser.  Name it accordingly, so that `git_revparse_mode_t`
is now `git_revspec_t`.  Deprecate the old name.

Only exclude tags files/folders and build folders at the repository root.

Our test resources shouldn't count toward our language statistics.  We
have some PHP checked in to the test resources, but I assure you that we
do not have any production code in PHP and we shouldn't be blamed for
the test data.

This change now updates the `lru_window` variable to match the current
file's MRU window. This makes it such that it doesn't always choose the
file that happened to come last in the list of window files, and instead
should now correctly choose the file with the least-recently-used one.

ci: don't use ninja on macOS

Ninja is not installed by default on the macOS machines; stop trying to
use it.  Instead use `make -j` which should be roughly equivalent in
performance but supported everywhere.