vutrinh274/duckdb_polars_benchmarking ? reverse-engineered prompt

Build me a small Python project that benchmarks DuckDB against Polars on TPC H data and lets me run the whole thing from one command per scale factor.

I want to generate real dbgen data at fixed sizes, then run all 22 queries against both engines, capture timings, and build a report of the results. It should work for scale factors 10, 50, 100, 200, and 465, and it should keep each run in its own folder so results never overwrite each other.

Please also include a tuned runner for larger datasets, since plain runs can hit memory issues. For DuckDB, add support for memory limit and temp directory settings. For Polars, make the lazy path use streaming execution so big runs do not thrash swap. If a query gets killed or hangs, I want that handled cleanly and recorded instead of the whole benchmark dying.

Make the setup simple with one command to install dependencies, and if needed, look up current docs online to match the latest DuckDB and Polars behavior.