Include ambiguous into UninhabitedType identity

[sterliakov]

Fixes #19641, but also reveals a test that was only passing by coincidence. This inference has never worked correctly:

from typing import Mapping, Never

d: dict[Never, Never]
def run() -> Mapping[str, int]:
    return d

Now - what is the desired behaviour here? I understand both "should be denied" and "should be allowed", but would prefer denying (as we do now). If everyone agrees, I will update the test case to remove xfail and add the error. Cc @ilevkivskyi as the original author (#16122)

[ilevkivskyi]

@sterliakov

Now - what is the desired behaviour here? I understand both "should be denied" and "should be allowed", but would prefer denying (as we do now).

Yeah, I think we should deny this (i.e. Never should be only handled specially if the ambiguous flag is set, otherwise regular subtyping rules should apply).

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[sterliakov]

Ough, yes, we definitely should deny that - only inline empty collection literals should receive special treatment. https://mypy-play.net/?mypy=latest&python=3.12&flags=strict&gist=2000dac04e592ad8b92012aa697cdbe6