Fix scoping for function annotations, decorators and base classes

[anjsimmo]

Steps

• Add a ChangeLog entry describing what your PR does.
• Write a good description on what the PR does.

Description

This is a follow up to PR #3497 Fix class scope issues in undefined-variable checker. The earlier PR fixed false positives in which function annotations and decorators were being incorrectly reported as unable to access the class scope due to being incorrectly treated as within the function. This PR fixes false negatives that were caused by still searching the function scope.

Issue #1082, while reported as a false positive for used-before-assignment is really due to a false negative for undefined-variable. Issue #3461 was partially addressed by PR #3497, and this PR fixes the false negative case where we were not reporting type hints that reference the function parameter self (there are still lots of other open type hint issues, but this fixes the specific case reported). Issue #3434 was due to skipping only a single scope level when checking a class's base class list, when there could in fact be an arbitrary number of nested list comprehensions in between. This logic has been moved to a per-scope check similar to the way function annotations and decorators are now checked.

To implement this fix, it was necessary to keep track of the frame/scope node in NamesConsumer, as this may be different to the immediate frame/scope of the variable being checked.

Type of Changes

	Type
✓	🐛 Bug fix

Related Issue

Close #1082, #3434, #3461

[coveralls]

Coverage decreased (-0.01%) to 90.686% when pulling 84209b5 on anjsimmo:fix-scope into 3d85973 on PyCQA:master.

[coveralls]

Coverage decreased (-0.005%) to 90.693% when pulling 84209b5 on anjsimmo:fix-scope into 3d85973 on PyCQA:master.

[Pierre-Sassoulas]

Looks really good, thank you.

[anjsimmo]

Thanks @Pierre-Sassoulas, although it seems GitHub only auto-closed #1082. Could you please also close #3434 and close #3461?