FixedSizeBinary filter benchmarks vary by substantially from run to run

> I see other runs of the benchmarks have the same pattern (with the same numbers for the fsb benchmark). Somehow the second run is always takes 66% longer... #7463.

Yeah, it is really weird. It would be great if someone could look into that.

I'll file a ticket

_Originally posted by @alamb in https://github.com/apache/arrow-rs/issues/7573#issuecomment-2956613295_

Start by reproducing the FixedSizeBinary filter benchmark and compare repeated runs, including the behavior reported in #7463. Trace the benchmark setup and execution path to identify why the second run is consistently about 66% slower; done means the cause is explained and benchmark results are stable or the variance is documented and addressed.

Written by the indexing model from the issue text.