Fixes #3762: Fix dataframe groupby aggregations when keys contain NaNs

[stress-tess]

This PR (fixes #3762) using dataframe groupby with keys that contain NaNs would cause the aggregations to fail. To resolve this, we mask out the values that belong to the NaN segment

Before this PR:

>>> df = ak.DataFrame({"A":[1,2,2,np.nan],"B":[3,4,5,6], "C":[1,np.nan,2, 3]})
>>> df.groupby("A").count()
---------------------------------------------------------------------------
ValueError
...
ValueError: Attempt to group array using key array of different length

After this PR:

>>> df = ak.DataFrame({"A":[1,2,2,np.nan],"B":[3,4,5,6], "C":[1,np.nan,2, 3]})
>>> df.groupby("A").count()
     B  C
A
1.0  1  1
2.0  2  1 (2 rows x 2 columns)

>>> df.to_pandas().groupby("A").count()
     B  C
A
1.0  1  1
2.0  2  1

>>> df.groupby(["A","C"],as_index=False).count()
     A    C  B
0  1.0  1.0  1
1  2.0  2.0  1 (2 rows x 3 columns)

>>> df.to_pandas().groupby(["A","C"],as_index=False).count()
     A    C  B
0  1.0  1.0  1
1  2.0  2.0  1

[ajpotts]

Looks good! Thanks for doing this one!