An S3-backed analytical dataset contains millions of tiny Parquet files. Queries spend substantial time listing and opening files despite selecting modest data volumes. Which remediation should the team evaluate?

1. Rename each file with a shorter basename without changing file count or content.
2. Create one additional empty file per existing file to increase parallelism indefinitely.
3. Compact files into appropriately sized outputs while preserving partition layout and safe table publication semantics.
4. Convert all Parquet data to uncompressed row-based text and keep the same tiny-file count.