結果 : python dataframe drop duplicates keep first