[WIP] Prod/Sum of all-NA / all-empty

[TomAugspurger]

To make review a bit easier, I've opened this before it's complete.

Of course we still need to decide

1. Do we want
   a. sum([]) = 0; prod([]) = 1
   b. sum([NaN]) = 0; prod([NaN]) = 1
2. Do we like the keyword approach (requiring people who want empty to be NA to always specify it) or do we want a new method?
3. Do we like the empty_is_na keyword? Alternatives are fill_value, ...
4. Does this apply to groupby (unobserved categoricals)
5. Does this apply to resample?

There are still some failing tests. I've tried to keep the commits separate:

• groupby: e96e386
• resample: 20f2ccc

closes #18678

[pep8speaks]

Hello @TomAugspurger! Thanks for submitting the PR.

• In the file pandas/tests/frame/test_analytics.py, following are the PEP8 issues :

Line 804:29: E225 missing whitespace around operator

• In the file pandas/tests/groupby/test_categorical.py, following are the PEP8 issues :

Line 686:49: W292 no newline at end of file

• In the file pandas/tests/test_nanops.py, following are the PEP8 issues :

Line 1058:48: W292 no newline at end of file

[TomAugspurger]

Superseded by #18876