Reduce false-positives emitted from Y026 TypeAlias check

[AlexWaygood]

Fixes #205.

I've temporarily renamed the check from Y026 to Y038 so that I can see the impact using typeshed_primer in CI. I'll revert the renaming once the CI has completed.

(EDIT: The output looked good, so I reverted the renaming.)

[AlexWaygood]

I've done some random skimming of the output from typeshed_primer here, and it looks like this gets rid of the vast majority of the false positives while retaining the vast majority of the true positives.

[AlexWaygood]

There are still some false positives, even with this patch. I'm working on it.

[JelleZijlstra]

I know it's a draft, but I want to clear my requested review list :)

I'm wary of a heuristic-heavy approach like this since if it goes wrong, it will be hard for new contributors to understand the problem.

What if we limit ourselves to extremely obvious cases like those where the RHS has a | at the top level?

[AlexWaygood]

I know it's a draft, but I want to clear my requested review list :)

I'm wary of a heuristic-heavy approach like this since if it goes wrong, it will be hard for new contributors to understand the problem.

What if we limit ourselves to extremely obvious cases like those where the RHS has a | at the top level?

I'm moving in that direction as well after discovering more false positives when working on python/typeshed#7630 :)

[AlexWaygood]

With the latest version of this patch (and following python/typeshed#7630), there are now only 4 hits in typeshed:

./stdlib/builtins.pyi:1265:1: Y026 Use typing_extensions.TypeAlias for type aliases
./stdlib/builtins.pyi:1384:1: Y026 Use typing_extensions.TypeAlias for type aliases
./stdlib/typing.pyi:1141:5: Y026 Use typing_extensions.TypeAlias for type aliases
./stdlib/typing.pyi:1153:5: Y026 Use typing_extensions.TypeAlias for type aliases

These are all places where we can't currently use TypeAlias due to some kind of bad interaction between PEP 604 and PEP 612 for mypy.