Simplify and improve Series.sum

[hamdanal]

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

Series.sum was annotated to return the type of the series elements except for Series[bool] that was special-cased to return int. This is conceptually wrong, although it works for the most common series types. The return value of Series.sum depends on the return value of its elements' __add__ method. Fixing this simplifies the signature of sum and makes it work generically (for example if in the future Series[np.bool] becomes a thing).

Note that I didn't add a test case for this as it is already tested in test_types_sum in test_series.py. Note also that the _SupportsAdd type-check-only protocol is added in the series module instead of _typing as it is very specific to this use case.

[Dr-Irv]

I just realized that you added the use of type_check_only in a previous PR, but now I'm wondering if it makes sense in a stub file, because if you're using a stub file, doesn't that mean that its contents are only for type checking?

[hamdanal]

This decorator is meant to be used with functions/classes that exist at type checking time but not at runtime. It is not strictly needed but it tells type checkers to flag any misuse of the decorated object at runtime. It is part of the typing module but it only exists at type checking time. The API documentation is here https://docs.python.org/3/library/typing.html#typing.type_check_only and the user guide is here https://typing.python.org/en/latest/guides/writing_stubs.html#stub-only-objects

[Dr-Irv]

OK - thanks for the explanation. We probably should be using that more in the pandas stubs

[Dr-Irv]

thanks @hamdanal

@cmp0xff you may want to take a look at this. It might inspire an idea to deal with the issues with TimestampSeries removal.