[DF] Handle NaN values properly from CSV

Initially reported in the forum here. RDataFrame breaks when some columns have NaN values, most noticeably when requesting a histogram of a column without NaN values.