BUG: groupby with dropna=False drops nulls from categorical groupers

[rhshadrach]

• closes BUG: Get wrong result when groupby category column with dropna=False #36327 (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

The function nunique here is faster than using pd.unique, but only works on integers:

arr = np.random.randint(0, 300, size) %timeit len(unique(arr)) %timeit nunique_ints(arr) size: 10 8.94 µs ± 133 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each) 2.94 µs ± 12.2 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each) size: 100 9.82 µs ± 22.2 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each) 3.01 µs ± 5.16 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each) size: 1000 11 µs ± 76.5 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each) 3.74 µs ± 9.7 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each) size: 10000 27.9 µs ± 66.8 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each) 10.4 µs ± 9.43 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each) size: 100000 187 µs ± 214 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each) 78.8 µs ± 408 ns per loop (mean ± std. dev. of 7 runs, 10,000 loops each) 

[mroeschke]

Merge conflict otherwise LGTM

[mroeschke]

Thanks @rhshadrach