Fix bug when narrowing union containing custom eq against custom eq

[hauntsaninja]

I rewrote the logic for custom equality in #20643 . This is a soundness bug in that rewrite.

Fixes #20750

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

pip (https://github.com/pypa/pip)
+ src/pip/_internal/resolution/resolvelib/candidates.py:293: error: Item "None" of "Link | None" has no attribute "is_wheel"  [union-attr]
+ src/pip/_internal/resolution/resolvelib/candidates.py:293: error: Item "None" of "Link | None" has no attribute "is_file"  [union-attr]
+ src/pip/_internal/resolution/resolvelib/candidates.py:294: error: Item "None" of "Link | None" has no attribute "filename"  [union-attr]
+ src/pip/_internal/resolution/resolvelib/candidates.py:305: error: Item "None" of "Link | None" has no attribute "is_wheel"  [union-attr]
+ src/pip/_internal/resolution/resolvelib/candidates.py:306: error: Item "None" of "Link | None" has no attribute "is_file"  [union-attr]

schema_salad (https://github.com/common-workflow-language/schema_salad)
+ schema_salad/tests/test_examples.py: note: In function "test_scoped_id":
+ schema_salad/tests/test_examples.py:395:23: error: Argument 2 to "makerdf" has incompatible type "int | float | str | CommentedMap | CommentedSeq | None"; expected "CommentedMap | float | str | CommentedSeq"  [arg-type]
+ schema_salad/tests/test_examples.py:408:23: error: Argument 2 to "makerdf" has incompatible type "int | float | str | CommentedMap | CommentedSeq | None"; expected "CommentedMap | float | str | CommentedSeq"  [arg-type]
+ schema_salad/tests/test_examples.py:421:23: error: Argument 2 to "makerdf" has incompatible type "int | float | str | CommentedMap | CommentedSeq | None"; expected "CommentedMap | float | str | CommentedSeq"  [arg-type]
+ schema_salad/tests/test_examples.py:434:23: error: Argument 2 to "makerdf" has incompatible type "int | float | str | CommentedMap | CommentedSeq | None"; expected "CommentedMap | float | str | CommentedSeq"  [arg-type]

[hauntsaninja]

Oh cool, for first time primer indicates we now actually do have a place where unsoundly special casing None could make sense

[nickdrozd]

Tested: works. Thanks!

[ilevkivskyi]

@hauntsaninja

Oh cool, for first time primer indicates we now actually do have a place where unsoundly special casing None could make sense

Do we have an issue about this so that we will not forget? Probably makes sense to do this before the next release, so that people will not see this as a perceived regression.

[hauntsaninja]

It is easy to make changes now, I will just open a PR for it

[hauntsaninja]

Something like this should do it: #20756

[hauntsaninja]

Oh and in general I am leaving some notes for myself in #20564
If there's anything you would like me to not forget feel free to post in there