how to filter PySpark dataframe using row number