docs(python): make an explicit note in `read_parquet` and `scan_parquet` about hive-style partitioning (point to `scan_pyarrow_dataset` instead) #3689