Analysts query a large S3 dataset in Athena, usually filtering by event date and selecting only a few columns. The data is stored as unpartitioned CSV. Which redesign most directly reduces unnecessary scanning?

1. Store compressed Parquet and partition by a useful event-date granularity aligned with query filters.
2. Create one partition for every unique row identifier even though queries do not filter by that identifier.
3. Compress the entire CSV dataset into one large non-splittable object without any date partitioning.
4. Add LIMIT 10 to every query and assume Athena will read only ten source rows for all query shapes.