Add Commit.is_shallow property; document stats() limitation at shallow boundary

[harshitayadavv]

Problem

Accessing .stats on a commit at the boundary of a shallow clone raises
git.exc.GitCommandError: fatal: bad object <sha>. This happens because
the commit object itself still references its parent's SHA, but that
parent was never fetched — it's beyond the shallow depth. There's currently
no way to detect this ahead of time without hitting the crash.

Reproduction

git clone --depth 5 <any repo with more history> repo
cd repo
python -c "
import git
repo = git.Repo('.')
oldest = list(repo.iter_commits())[-1]
print(oldest.parents) # non-empty -- references a SHA that doesn't exist locally
oldest.stats # raises GitCommandError: fatal: bad object <sha>
"

Fix

Adds a Commit.is_shallow property that checks whether the commit's hexsha
appears in the repository's shallow file -- a cheap, local check with no
git subprocess call. Also documents this limitation in the stats
docstring, pointing to is_shallow.

Includes a test (test_is_shallow) that reproduces the failure via a real
local shallow clone (using no_local=True, since Git ignores --depth for
local-optimized clones) and confirms both that the boundary commit reports
is_shallow == True and that .stats raises as expected on it.

Ran the full test suite locally (Windows, Python 3.11): 660 passed, 9
unrelated pre-existing failures (Windows symlink privilege requirements,
a missing-remote assumption in one tutorial test, and PATH resolution in
shell-impostor tests) -- none touching commit.py or stats.

AI disclosure

Per CONTRIBUTING.md: I used Claude to help design, implement, and test
this change. I reviewed, ran, and verified every step myself, including
reproducing the underlying bug firsthand before writing the fix.

[Byron]

Thanks for giving this a shot.

Can we reduce this to just the documentation change for a quick merge? Otherwise, the .is_shallow() should rather be on Repo to support testing multiple commits at the same time. I wouldn't want to encourage per-commit calls given they are IO-heavy then. Alternatively, offer a way to get a set of hashes that are the shallow boundary so users can do their is-contained checks themselves.

[harshitayadavv]

@Byron Done, reduced to just the docstring note on stats(). Happy to follow up separately with a Repo-level version (e.g. a method returning the set of shallow-boundary hashes) if that'd be useful.