Fixes python#3315

Co-authored-by: hauntsaninja <>

Refs #5710

Adds special-case handling for raising `NotImplemented`:
```python
raise NotImplemented  # E: Exception must be derived from BaseException; did you mean "NotImplementedError"?
```

Per the linked issue, there's some debate as to how to best handle
`NotImplemented`. This PR special-cases its behavior in `raise`
statements, whereas the leaning in the issue (at the time it was opened)
was towards updating its definition in typeshed and possibly
special-casing its use in the relevant dunder methods.

Going the typeshed/special-dunder route may still happen, but it hasn't
in the six years since the issue was opened. It would be nice to at
least catch errors from raising `NotImplemented` in the interim.

Making this change also uncovered a regression introduced in
python/typeshed#4222. Previously, `NotImplemented` was annotated as
`Any` in typeshed, so returning `NotImplemented` from a non-dunder would
emit an error when `--warn-return-any` was used:
```python
class A:
    def some(self) -> bool: return NotImplemented  # E: Returning Any from function declared to return "bool"
```
However, in python/typeshed#4222, the type of `NotImplemented` was
updated to be a subclass of `Any`. This broke the handling of
`--warn-return-any`, but it wasn't caught since the definition of
`NotImplemented` in `fixtures/notimplemented.pyi` wasn't updated along
with the definition in typeshed. As a result, current mypy doesn't emit
an error here
([playground](https://mypy-play.net/?mypy=1.11.2&python=3.12&flags=strict%2Cwarn-return-any&gist=8a78e3eb68b0b738f73fdd326f0bfca1)),
despite having a test case saying that it should. As a bandaid, this PR
add special handling for `NotImplemented` in return statements to treat
it like an explicit `Any`, restoring the pre- python/typeshed#4222
behavior.