Consistent Naming Standard When Using Dict in GroupBy .agg

[WillAyd]

• closes API: inconsistencies between grouped and grouped[['col']] in groupby #21790
• [] tests added / passed
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

This is a WIP as all of the tests in the groupby directory pass, but there are some outside of that which need touch ups. This is a relatively aggressive change so I figured I'd get buy in before going outside of that to modify tests.

This not only addresses the referenced issue but provides consistent expectations around columns returned when using a dict argument in a DataFrameGroupBy, namely that the returned object will have a MultiIndex where the top level is the column name and the subsequent level is the aggregation performed.

To play devil's advocate, this may be considered undesirable as it now returns a MultiIndex in some cases where that was flat before. However, I'd counter that this:

• Behavior is more consistent and easier to communicate
• Provides a return value whose column labeling is less ambiguous AND
• Simplifies the code base, providing us a clear path forward for whatever the renaming solution actually SHOULD be

Curious to hear other's feedback @jreback @TomAugspurger @jorisvandenbossche

[pep8speaks]

Hello @WillAyd! Thanks for submitting the PR.

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

Line 210:1: E302 expected 2 blank lines, found 1

[WillAyd]

FYI plan to move this to the aggregate sub-directory of tests

[WillAyd]

Closing as this should be part of larger discussion