Standardize performance benchmark

As we're separating algorithmic changes from performance changes, there should be a standardized way to measure how two different versions perform from a purely performance point of view -- rate, time for full run, etc.

No file, test, or entry point is named. Start by surveying the repository for the recommendation algorithm's execution entry points and any existing timing or evaluation tests. Done means documenting and implementing an agreed standard for comparing versions by rate, full-run time, and other performance metrics.

Written by the indexing model from the issue text.