use nullable integer dtypes

[SteffenMeinecke]

In several cases nullable integer dtypes can avoid falling from integers to floats if None-values appear. At least with #1624, this issue must be handled. Missing support of pd.NaN in numpy must be taken into account.