[TASK] Benchmark PERFILE deletion vector reader

[razajafri]

We need to benchmark the perfile reader to document if we see any performance regression.

There are multiple steps involved to achieve this

• Convert the sf100 parquet data to Delta Lake format
• Create two sets of tables, one with 2% of the data deleted, and the other with 50% of the data deleted
• Benchmark the query against branch-25.04 and the feature branch

Once documented, the above steps should be repeated with sf3k if needed