B042: don't count a parameterized generic base as an exception super class

[Eljees]

Refs #538.

B042 treats a class as an exception when its name looks like one and it has any base at all, so class FailOnException(ContextManager[T]) from the issue is checked like an exception class. Resolving real inheritance from BaseException is out of reach for a single-file check, so this takes the narrow, syntactic part of it: a parameterized generic base, such as ContextManager[T] or typing.Iterator[int], no longer counts as the super class that makes a class named ...Exception or ...Error an exception. A generic base whose own name looks like an exception, such as MyError[int], is now recognised, which it wasn't before because only plain names were checked.

Everything else is unchanged. A class named like an exception with any non-generic base is still checked, since that base may be an exception under another name, and bases named like exceptions are matched as before.

New cases in tests/eval_files/b042.py:

• FailOnException(typing.ContextManager[int]) and FailOnError(typing.Iterator[int]): no longer reported. Both are reported on main;
• MyGenericError(typing.Iterator[int], Anything): still reported, through the non-generic base;
• Anything5(MyError[int]): now reported. main misses it.

On main, the eval test fails on exactly these lines. With the change, pytest tests: 81 passed, 2 skipped. There is a changelog entry under UNRELEASED.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[cooperlees]

This is muse (muse-spark, posting on behalf of @cooperlees via his coding agent).

Reviewed the diff and verified locally: checked out the PR head (a0859da) and ran the full suite in an isolated venv in /tmp — 83 passed, including the new b042.py eval cases.

What I checked:

• FailOnException(typing.ContextManager[int]) / FailOnError(typing.Iterator[int]): correctly no longer flagged — the only bases are ast.Subscript, so concrete_bases is empty and the unwrapped Attribute base doesn't match the exception-name heuristic. This fixes the false positive in #538.
• MyGenericError(typing.Iterator[int], Anything): still flagged via the non-generic base — good, preserves the "base may be an exception under another name" behavior.
• Anything5(MyError[int]): now flagged by unwrapping the Subscript to Name(MyError) — a genuine improvement over main.
• Plain cases (class FooError(Exception), bare-name bases) are untouched; the change is narrowly scoped to ast.Subscript bases.

One non-blocking observation: an attribute-style generic base like typing.MyError[int] still won't match since only ast.Name is checked after unwrapping — but that's a pre-existing gap (non-subscripted Attribute bases were never matched either), so keeping this PR narrow is the right call.

Changelog entry under UNRELEASED looks good. Approving.

[cooperlees]

Many thanks !