[nlsfnr]

fixes #13926 and #16992 (duplicates), i.e. false possibly-undefined errors after exhaustive match statements.

This PR applies the same approach as #20149, which did the same for if statements.

Adds a new attribute MatchStmt.is_exhaustive, which is set by the type checker when the remaining subject type is empty after the last case. This allows partially_defined.py to skip the implicit fall-through branch, which it already does for trailing catch-all cases.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[chrikrah]

@nlsfnr nothing blocks this from my side. Like #20149, a constrained TypeVar still reports: expand_typevars() checks copies, so the original MatchStmt keeps is_exhaustive=False.

# git archive fee4b07, Python 3.12, mypy 2.5.0+dev resolves into the export
$ python -m pytest -q -n 4 mypy/test/testcheck.py -k "PossiblyUndefined or Match"
264 passed, 1 skipped        # head fee4b07
$ python -m pytest -q mypy/test/testcheck.py -k testExhaustiveMatchNoFallThrough
1 failed        # partially_defined.py skip_branch() removed: lines 15, 23, 31 unexpectedly report
$ python -m mypy --no-incremental --enable-error-code possibly-undefined t.py
t.py:11: error: Name "y" may be undefined  [possibly-undefined]    # f(x: T): match, case int() / case str()
t.py:26: error: Name "y" may be undefined  [possibly-undefined]    # h(x: T): if isinstance int / elif str
# g(x: int | str), the same match as f: clean
# not run: mypyc build, self-check

non-blocking: a separate issue?