Document overall perf process

The perf process is in flux, but that's even more reason to document that latest state of things. We currently only really document the triage process, but the perf management has grown to be more than triage.

I wrote a bit about this on Zulip:

Essentially the perf-regression label is used to indicate that the latest known performance state of the PR is a regression. perf-regression-triaged is the label applied when something has been done to address the performance issue. This can be one of several things:

• An issue has been created to address the performance issue (linking to the original PR)
• The author and reviewer have determined that the change in the PR is worth the regression (e.g., a correctness fix - better to be slow and correct than fast and wrong)
• The regression isn't actually a regression (i.e., there's something wrong with perf.rlo) - this should be accompanied by an issue on rustc-perf
• Another PR has been opened that fixes the perf regression (with a link)

Review the existing documentation for the performance triage process and the Zulip guidance quoted in the issue. Document the broader performance-management workflow, including how perf-regression and perf-regression-triaged are resolved and what links or follow-up issues are expected; done means the current process is described beyond triage.