Remove redundant .fillna(0) at the end of groupby.size()/.count()

[dchigarev]

Currently, each groupby.size() call is finished with .fillna(0):

modin/modin/pandas/groupby.py

Line 899 in 632d724

return result.fillna(0)

The behavior was originally introduced in #1802 in order to match pandas in the sense of not having NaN values in the result of .size()/.count(). The NaNs appeared at the reduction phase of groupby due to some specific pandas behavior:

>>> import pandas as pd # pandas 1.0.5
>>> df = pd.DataFrame({"a": [1, 2], "b": [3, 4], "c": [4, 3]}).astype({"a": "category"})
>>> df.groupby(["a", "b"]).sum()
       c
a b
1 3  4.0
  4  NaN
2 3  NaN
  4  3.0

Running the same code on the latest pandas gives the expected behavior (not producing NaNs), meaning that we now can remove the finalizing .fillna(0) in our modin's code

>>> import pandas as pd # pandas 1.5.3
>>> df = pd.DataFrame({"a": [1, 2], "b": [3, 4], "c": [4, 3]}).astype({"a": "category"})
>>> df.groupby(["a", "b"]).sum()
     c
a b
1 3  4
  4  0
2 3  0
  4  3