Hey @Colin Ho seeing some odd behavior that is causing the codspeed noise on github. when I run

DAFT_RUNNER=native pytest -m benchmark tests/benchmarks/test_interactive_reads.py -k "test_iter_rows_first_row" --benchmark-min-rounds=50

and increase the number of rounds, we seems to get higher and higher std dev, which isn't a good sign. I think the rounds are affecting each other and it's not being cleaned up