[ty] Reject fixed tuples for non-inferable TypeVarTuples

[carljm]

On main, a function returning tuple[*Ts] can wrongly return an unrelated fixed-length tuple without an error. For example, returning () from a variadic-identity function caused ty to infer a nonempty tuple at call sites even though the runtime result was an empty tuple.

Only expand a non-inferable TypeVarTuple target when the source tuple is variable-length. Fixed tuples now reach the ordinary type-variable rejection instead of repeating the same tuple comparison and succeeding through the recursion guard. Inference and explicit constraint assumptions keep their existing behavior.

Depends on #27950 to avoid cascading errors after a missing TypeVarTuple unpack, and transitively on #27946 to preserve exhaustiveness for variadic class patterns.

Fixes astral-sh/ty#4342.

Test plan

Added mdtests for empty and nonempty fixed returns, annotated assignments, matching tuple prefixes and suffixes, subtyping, and fixed tuples containing Any or Never. Constraint-implication coverage checks that an applicable assumption permits the relationship while incompatible elements and unconstrained packs remain rejected.

[astral-sh-bot]

Typing conformance results

No changes detected ✅

Current numbers

The percentage of diagnostics emitted that were expected errors held steady at 97.68%. The percentage of expected errors that received a diagnostic held steady at 93.44%. The number of fully passing files held steady at 109/136.

[astral-sh-bot]

Memory usage report

Memory usage unchanged ✅

[astral-sh-bot]

ecosystem-analyzer results

No diagnostic changes detected ✅

Full report with detailed diff (timing results)

[dhruvmanila]

This does raise a question on what to do for narrowing like:

def reject_empty[*Ts](values: tuple[*Ts]) -> tuple[*Ts]:
    if len(values) == 0:
        # error on this branch but should it?
        return values
    return ()  # error: [invalid-return-type]

Both Pyright and mypy don't error on the first return statement, pyrefly errors on both return statements.

This can also be fixed in follow-up but it would be useful to have a test case for it.