Add page index and content-defined chunking to reshard script

reshard_parquet.py

@@ -54,7 +54,7 @@ def reshard(shards: list[Path], split: str, rows_per_shard: int, out_dir: Path)
             remainder = merged.slice(rows_per_shard)
 
             name = f"{split}-{out_idx:05d}-of-{num_out:05d}.parquet"
-            pq.write_table(to_write, out_dir / name, compression="zstd")
+            pq.write_table(to_write, out_dir / name, compression="zstd", row_group_size=200_000, write_page_index=True, use_content_defined_chunking=True)
             out_idx += 1
             print(f"    wrote {name} ({to_write.num_rows:,} rows)")
 
@@ -70,7 +70,7 @@ def reshard(shards: list[Path], split: str, rows_per_shard: int, out_dir: Path)
         import pyarrow as pa
         merged = pa.concat_tables(buf_rows)
         name = f"{split}-{out_idx:05d}-of-{num_out:05d}.parquet"
-        pq.write_table(merged, out_dir / name, compression="zstd")
+        pq.write_table(merged, out_dir / name, compression="zstd", row_group_size=200_000, write_page_index=True, use_content_defined_chunking=True)
         out_idx += 1
         print(f"    wrote {name} ({merged.num_rows:,} rows)")