Remove old shard-named files

Rename validation shard to validation.parquet

Rename train shard to train.parquet

Remove old 16-shard Parquet files

Single shard per split with page index for random access

Re-shard with page index and smaller row groups for dataset viewer compatibility

Remove old 307-shard Parquet files

Re-shard to 16 larger Parquet files (1M rows each)

Convert JSONL to sharded Parquet for dataset viewer compatibility