coalesce - pandas version shouldn't replace NaN when other is also NaN-like

[machow]

For example, in the code below, the output replace np.nan from the first series, with None...

x = pd.Series([1., None, None])
abc = pd.Series(['a', 'b', None])
coalesce(x, abc)

#        0    1
#       1    b
#        2    None
#        dtype: object