Review expected failures

[Dr-Irv]

In #213 @twoertwein marked typing tests that we expect to fail using TYPE_CHECKING_INVALID_USAGE . The idea here is that we are testing code that we want the type checkers to pick up as being incorrect. In many cases, pyright does figure that out. mypy figures it out sometimes, and there are occasions where both type checkers see what we expect to fail as "correct" code.

At some point in time in the past, when the stubs were in the Microsoft repository, those tests were doing the "right" thing, i.e., both type checkers were marking those tests as "failures", and the # type: ignore comments were correct, in the sense that the type checkers saw that they were correct ignores.

Now, possibly due to changes in the settings used for mypy and pyright, as well as evolutions in the type checkers themselves, those tests are no longer failing as we expect them to. Some investigation is required to determine whether this is due to something changing in the stubs, or to particular settings in the type checkers.

[twoertwein]

I think why

pandas-stubs/tests/test_timefuncs.py

Line 208 in 417c3d0

ssum: pd.Series = s1 + s2 # TODO both: ignore[operator]

works might be because of this overload

pandas-stubs/pandas-stubs/core/series.pyi

Line 1163 in 417c3d0

self, other: num | _str | Timedelta | _ListLike | Series[S1]

Would need type ignores but it might work to have an overload before this fallback overload of TimestampSeries + TimestampSeries -> NoReturn. This would overlap with the last overload (ignore in pandas-stubs) but it should be correctly parsed by mypy/pyright when testing external code as type checkers will use the first matching overload.