Fix incompatible behaviour with pandas for floordiv with np.nan

[itholic]

Resolves #1428

>>> pser = pd.Series([-100, 0, 100, None, np.nan], name="Koalas")
>>> kser = ks.from_pandas(pser)

>>> pser // np.nan
0   NaN
1   NaN
2   NaN
3   NaN
4   NaN
Name: Koalas, dtype: float64

>>> kser // np.nan                                                                                                                
0   NaN
1   NaN
2   NaN
3   NaN
4   NaN
Name: Koalas, dtype: float64

[codecov-io]

Codecov Report

Merging #1429 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master    #1429   +/-   ##
=======================================
  Coverage   95.14%   95.14%           
=======================================
  Files          34       34           
  Lines        7967     7967           
=======================================
  Hits         7580     7580           
  Misses        387      387           

Impacted Files	Coverage Δ
databricks/koalas/base.py	97.64% <ø> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 83e44bf...5c96153. Read the comment docs.

[HyukjinKwon]

LGTM otherwise.