Efficiently groups Parquet files based on their schema, allowing for the detection of schema inconsistencies between files on different filesystems. Achieves high efficiency by reading metadata without loading entire files.