Document performance measurements

[andyleiserson]

I posted the following metrics in the chat a while back:

OPRF IPA for 32,768 records without any vectorization: 118s OPRF IPA for 32,768 records with vectorization of share conversion, quicksort, and aggregation: 5.7s (~20x speedup)

The best numbers have some changes that are not merged: https://github.com/andyleiserson/ipa/commit/b238faec904ee557c4a8f0ea3bec8c5dcf12c0f8

There are a few outstanding tasks here:

• Finish the buffer tunability changes, to use instead of TotalRecords::Indeterminate
• Investigate the tracing performance issues
• Make sure our performance is tuned for the real-world HTTP network case, not the in-memory network case
• Document the canonical performance measurement commands, perhaps in https://github.com/private-attribution/ipa/blob/main/ipa-core/benches/README.md.