結果 : spark dataframe drop duplicates keep first