BUG: groupby doesn't identify null values when sort=False

[rhshadrach]

Closely related to #48476

As far as I can tell this only occurs when the input dtype to groupby is object.

df = pd.DataFrame({'a': [np.nan, pd.NA, None], 'b': [1, 2, 3]})
gb = df.groupby('a', sort=True, dropna=False)
print(gb.sum())

#      b
# a     
# NaN  6

but with sort=False:

df = pd.DataFrame({'a': [np.nan, pd.NA, None], 'b': [1, 2, 3]})
gb = df.groupby('a', sort=False, dropna=False)
print(gb.sum())

#       b
# a      
# NaN   1
# <NA>  2
# None  3

I think we should prefer the sort=True behavior as that is the default value for now, but prefer sort=False in the long run.

[rhshadrach]

This is a regression that occurred in #46601, cc @rhshadrach. Prior to this, the sort=False behavior agreed with sort=True in collapsing different types of null values.

[phofl]

Added 1.5 milestone