Regression: false positive about overload item never being matched

[JukkaL]

This fragment generates a false positive, starting from #15913:

from typing import overload, Any

class C:
    @overload
    def f(self, e: str, **kwargs) -> str: ...
    @overload
    # error: Overloaded function signature 2 will never be matched: signature 1's
    #        parameter type(s) are the same or broader
    def f(self, *args, **kwargs) -> Any: ...
    def f(self, *args, **kwargs):
        pass

The signature of item 2 is wider than item 1 due to accepting an arbitrary number of positional arguments and with arbitrary types, so this is a false positive.

cc @ilevkivskyi

[ilevkivskyi]

This is one of those case when I ask myself how did this ever work, and it turns out it didn't. It turns out having (*args: Any, **kwargs: Any) as a fallback overload caused false positives almost always. So this is not really a regression, now mypy is consistently broken, LOL.

Anyway, the fix is in #16343, it turns out we never propagated is_proper_subtype flag to is_callable_compatible(). This is a major change, let's see how bad the fallout will be.