Fine tune flaky resampling peakmem benchmark

[vasil-pashov]

Reference Issues/PRs

What does this implement or fix?

Peakmem params are tuned for the machine that runs the tests. It has 16 CPU threads and 24 IO threads. Having too much segments on leads to variability of ~20% in the processing pipeline because of the scheduling. 3_000_000 rows are 30 segments, a bit more than the number of IO threads but still not enough to cause large variability

Any other comments?

Checklist

Checklist for code changes...

• Have you updated the relevant docstrings, documentation and copyright notice?
• Is this contribution tested against all ArcticDB's features?
• Do all exceptions introduced raise appropriate error messages?
• Are API changes highlighted in the PR description?
• Is the PR labelled as enhancement or bug so it appears in autogenerated release notes?

[poodlewars]

How have you tested that this reduces the variability?

Needs a rebase - it clashes with my change that removes to 20M rows case.

[vasil-pashov]

How have you tested that this reduces the variability?

I had a script that ran the benchmark 1000 times and gathered the lowest and highest reported memory.