• closes Inconsistent behavior when groupby pandas Categorical variables #31422
• closes BUG: df.groupby().count() returns NaN instead of Zero #35028
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

Behavioural Changes
Fixing two related bugs: when grouping on multiple categoricals, .sum() and .count() would return NaN for the missing categories, but they are expected to return 0 for the missing categories. Both these bugs are fixed.

Tests
Tests were added in PR #35022 when these bugs were discovered and the tests were marked with an xfail. For this PR the xfails are removed and the tests are passing normally. As well, a few other existing tests were expecting sum() to return NaN; these have been updated so that the tests now expect to get 0 (which is the desired behaviour).

Pivot
The change to .sum() also impacts the df.pivot_table() if it is called with aggfunc=sum and is pivoted on a Categorical column with observed=False. This is not explicitly mentioned in either of the bugs, but it does make the behaviour consistent (i.e. the sum of a missing category is zero, not NaN). One test on test_pivot.py was updated to reflect this change.

Default Behaviour
Because df.groupby() and df.pivot_table() have observed=False as the default, the default behaviour will change for a user calling df.groupby().sum() or df.pivot_table(..., aggfunc='sum') if they are grouping/pivoting on a categorical with missing categories. Previously the default would give them NaN for the missing categories, now the default will give them 0.

What is the appropriate to highlight/document this change to the default behaviour?