Operators execution benchmarking

[connortsui20]

I think we want a consolidated way to measure execution speed of any arbitrary array, which imo is more complicated than just setting up a simple execution context and running execute a bunch of times, we want more fine-grained detail.

We want to know how long it takes for optimization, what rules get applied, how long each step of execution takes, etc. I'd rather deal with this separately and get rid of this now so we don't have a bunch of codspeed regressions/improvements that don't actually mean anything.

Originally posted by @connortsui20 in #6146 (comment)

[connortsui20]

It's pretty unfortunate that I can't "reference in new discussion" instead of an issue with our new system of discussions...

[connortsui20]

#6146 will remove the bitpacking filter, the run end filter, and the pco filter benchmarks. So those need to be added back eventually.

cc @joseph-isaacs