[pyflakes] Fix false positive for __class__ in lambda expressions within class definitions (F821)

[danparizher]

Summary

Fixes #20562

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[ntBre]

Thank you! The tests look good, but what do you think about moving the new code to visit_expr to mirror the function version a bit more closely?

[danparizher]

Yeah, that approach makes sense to me. Thanks for the review!

[ntBre]

Hmm, it looks like there are some new false positives in the ecosystem report. I was hoping it was something simple with the scope popping, which is why I pushed a commit, but that didn't end up helping. We're flagging a case like this with F401:

import uuid


class C:
    uuid = lambda: str(uuid.uuid4())

and I don't fully understand why. We'll need to figure it out before we can merge this, though. Would you mind taking a look?

[danparizher]

The issue was that the DunderClassCell scope was being added during lambda definition processing instead of during lambda body analysis, which caused false positives when lambdas within classes referenced module-level variables like uuid.

I fixed it by moving the scope logic from visit_expr to visit_deferred_lambdas (similar to how function definitions work) and changing the scope detection to check the entire scope hierarchy rather than just the current scope, ensuring __class__ is accessible in lambdas within classes while preventing false positives for other variables.

[ntBre]

Ah okay, I guess my earlier suggestion about moving it to visit_expr was wrong then. Thank you for debugging and sorry for the bad suggestion!