use UnknownSeries in series.pyi

[MarcoGorelli]

ref #1171

• Closes #xxxx (Replace xxxx with the Github issue number)
• Tests added: Please use assert_type() to assert the type of any return value

These all look like cases where:

• value-dependent behaviour makes narrowing the inner type untractable (e.g. in align, 'int64' might get upcast to 'float64' depending on the values, which aren't known statically)
• another generic is instantiated, like in Rolling. .rolling_sum. Currently these all return 'float64', but it's not clear to me that they should? For nullable types, for example, 'int' could be preserved? Given that this might be considered a bug in pandas, I think just keeping UnknownSeries is safest for now? pandas issue for reference: BUG(?): rolling sum with pyarrow types results in float64 instead of preserving integer type pandas#61144

[Dr-Irv]

I think we may want to add a test like this (maybe more tests - not sure):

def test_unknown() -> None:
    s = pd.Series(pd.NA)

    def foo(sf: pd.Series):
        pass

    foo(s)

    check(assert_type(s + pd.Series([1]), pd.Series), pd.Series)

Idea is to make sure that an UnknownSeries can be passed and that arithmetic works. I'm fine for now if you just add the above.

Eventually, we need to deal with this comment in a PR: #1093 (comment) (and other comments in that PR).

But the incremental approach is fine. Would be interesting to convert all references in the stubs from plain Series to UnknownSeries.

[Dr-Irv]

thanks @MarcoGorelli