[flake8-pyi] Fix several correctness issues with custom-type-var-return-type (PYI019)

[AlexWaygood]

Summary

This PR fixes several correctness issues in our PYI019 rule:

• The rule would panic on this snippet. There shouldn't even be a diagnostic on this snippet, since it's not a valid use of the S type variable:

  class F:
      @classmethod
      def m[S](cls: type[S], /) -> S[int]: ...

• The rule would panic on this snippet. Again, there shouldn't even be a diagnostic emitted here; it's an invalid use of a type variable:

  class F:
      def m[S](self: S) -> S[int]: ...

• The rule had a false positive on this snippet, where type is shadowed locally:

  def shadowed_type():
      type = 1
      class A:
          @classmethod
          def m[S](cls: type[S]) -> S: ...

• The rule had a false negative on this snippet, where the fully qualified version of builtins.type rather than the implicit builtin binding:

  import builtins
  
  class UsesFullyQualifiedType:
      @classmethod
      def m[S](cls: builtins.type[S]) -> S: ...  # PYI019

• The rule had a false negative on this snippet, where the return annotation is type[S] rather than simply S:

  class Foo:
      @classmethod
      def f[S](cls: type[S]) -> type[S]: ...

Fixes #15849

Test Plan

I've added new fixtures added that trigger panics and/or incorrect behaviour on main.

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[ntBre]

This looks great! I had to throw in one nit, but the changes were easy to follow, especially with the new variable names, and the tests made it clear what was going on.