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.

When a branch is specified to check out in clone, update the remote
tracking `HEAD` to point to it.  This mimics git's behavior, when
`git clone -b <name>` is used.

This change stops using the seek+read/write combo to perform I/O with an
offset, since this is faster by one system call (and also more atomic
and therefore safer).

The init.defaultbranch option may be set, but empty.  In this case, we
should ignore it instead of trying to set our default branch to
`refs/heads/`.

If a symbolic reference points to something invalid, then do not try to
update it.

When HEAD points to a nonexistent or invalid branch - for example, to
`refs/heads/` - the fetch should be permitted to continue, but we should
not use it when creating the `for merge` option in the FETCH_HEAD file.

(This emulates git's behavior.)

Adds info about initializing options with git_blob_filter_options_init

The `git_blob_filter_options_init` function should be included, to allow
callers in FFI environments to let us initialize an options structure
for them.

Removes doc comment on `git_blob_filter_options.version`, moves
information to `git_blob_filter_options` doc comment to remain
consistent with other options structures' documentation.

`git_blob_filter_options_init` still needed; should be added in another
commit/PR (it's out of the scope of this PR, #5759), update this
documentation again.

Resolves #5756

    * Use pread/pwrite to avoid updating position in file descriptor
    * Emulate missing pread/pwrite on win32 using overlapped file IO

    * Emulated mmap based write without pagefault handling is not
    possible since IO happens outside of call to mmap and data is
    written to mapped memory
    * Potential emulation using userfaultfd() might be possible

Continue the zlib stream as long as we can make progress; stop when we
stop getting output _or_ when zlib stops taking input from us.

Fixes #5280

worktree_dir isn't validated when it should be

Without this, mbedTLS installs in non-default install locations
that are otherwise found by the `FindmbedTLS.cmake` module are not
found by the C preprocessor at compile time.

Update ntlm to include an htonll that is not dependent on system
libraries.

This change avoids using the `(void)0` construct for some of the mutex
`#define`s, since that makes the "return type" of those "functions" to
be `void` instead of `int`.

This should allow folks that build in non-thread-safe environments to
still be able to build the library.

Fixes: #5663

Since we're using atomic primitives to read and write into the config
map cache, we need to read/write something pointer-sized.  Use an
`intptr_t` for the config map cache.

Move `git_reference__is_valid_name` to `git_reference__name_is_valid`,
which returns errors and sets an out boolean parameter.

Cherry-pick `git__add_int64_overflow` functionality introduced in
e99e833f.

ci: add manual dispatch event for 1.1 branch

ci: use GITHUB_ENV instead of set-env for v1.1 branch

libgit2 v1.1.0

Update PCRE to 8.44

clone: update origin's HEAD

Improve the support of atomics

Instead, globally initialize the system page size.

This change:

* Starts using GCC's and clang's `__atomic_*` intrinsics instead of the
  `__sync_*` ones, since the former supercede the latter (and can be
  safely replaced by their equivalent `__atomic_*` version with the
  sequentially consistent model).
* Makes `git_atomic64`'s value `volatile`. Otherwise, this will make
  ThreadSanitizer complain.
* Adds ways to load the values from atomics. As it turns out,
  unsynchronized read are okay only in some architectures, but if we
  want to be correct (and make ThreadSanitizer happy), those loads
  should also be performed with the atomic builtins.
* Fixes two ThreadSanitizer warnings, as a proof-of-concept that this
  works:
  - Avoid directly accessing `git_refcount`'s `owner` directly, and
    instead makes all callers go through the `GIT_REFCOUNT_*()` macros,
    which also use the atomic utilities.
  - Makes `pool_system_page_size()` race-free.

Part of: #5592