This commit fixes the following issues:

1. In `git_revwalk__push_commit`, if `opts->insert_by_date` is true, `git_commit_list_insert_by_date` is called. However, by this point the commit wasn’t parsed yet, so the `time` field still has 0 as value. Solved by parsing the commit immediately if `opts->insert_by_date` is true.

2. In the same function, there was an error in the boolean logic. When `opts->insert_by_date` was true, the commit would be inserted twice, first “by date” (not really, due to the issue described above) and then in the first position again. Logic was corrected.

3. In `prepare_walk`, when processing `user_input` and building the `commits` list, the order was being inverted. Assuming both fixes above, this would mean we would start negotiation by the oldest reference, not the newest one. This was fixed by producing a `commits` list in the same order as `user_input`.

The output list for the list of “have” statements during the negotiation is still not the same as core git’s because there is an optimization missing (excluding ancestors of commits known to be common between the client and the server) but this commit brings it much closer to core git’s. Specifically, the example on the #6532 issue description now fetches exactly the same objects than core git, and other examples I tested as well.

repo: don't allow repeated extensions

Introduce Schannel and SSPI for Windows

v1.6.3

Add support for SSPI on Windows, which offers NTLM and Negotiate
authentication.

Add functions to use convert a string with length, instead of assuming
NUL termination.

In addition, move the utf8 to 16 conversion routines into the `git_utf8`
namespace instead of using namespaceless `git__` prefixed names.

Name the GSSAPI and ntlmclient authentication providers as such. Today
they're named after the authentication mechanism ("Negotiate", "NTLM")
instead of their implementation.

If we have competing implementations for the same mechanism (eg, a
future Windows SSPI-based provider for Negotiate and NTLM) then this
will get confusing.

Provide a stream interface for Schannel - the native crypto APIs - on
Windows. This allows Windows to use the same HTTP transport that all the
other platforms use, with its own native crypto.

Ultimately this allows us to deprecate WinHTTP and we need not add
support for our socket changes in two places (our HTTP stack and the
WinHTTP stack).

pack: use 64 bits for the number of objects

It is a bit too hidden why we want 64 bits when we're assigning to a 32-bit
integer later on to store the number of objects, so go back to uint32_t and cast
to uint64_t during the size calculation.

Keeping it as a 32-bit value means the min size calculation overflows or gets
truncated which can lead to issues with large packfiles.

cmake: don't include `include/git2`

Including the `include/git2` build path is a seemingly unnecessary
oversight to include the generated `experimental.h` file.

These declaration poses problems on some embedded or retro Linux systems
that deliberately disable support for wchar_t from their libc.

When the contents of an entire new directory is ignored, and `FORCE` is
specified to `git_index_add_all`, ensure that we expand the entire file
list. By default, diff will coalesce a fully ignored folder into a
single diff entry; expand it.

Ensure that when all files beneath a directory are ignored that we
add the files when FORCE is specified.

Ensure that when all files beneath a directory are ignored that we
actually ignore the files.

odb: restore `git_odb_open`

fs_path: let root run the ownership tests

Remove unused wditer variable

`git_odb_open` was erroneously removed during a refactoring; add it
back.

Found this randomly while debugging #6517

Revert #6503

The library name is correctly libgit2 (not git2) or libgit2-experimental
depending on configuration.

`git_odb_backend_loose` was erroneously removed during a refactoring;
add it back.

The `git_fs_path_owner_is_current_user` expects the root dir on unix
(`/`) to be owned by a non-current user. This makes sense unless root
(or euid == 0) is running the tests, which often happens during distro
build / packaging scripts. Allow them to run the tests.

This reverts commit 43e84e24.

This reverts commit f68b40c0.

remote: always populate old id in update tips

In b1e83cca we erroneously stopped
setting the old ID to zero; correct that.

If a user attempts to add a custom extension that the system already
supports, or that is already in their list of custom extensions, de-dup
it.