perf-regression: collect report at each staging release

The idea is to start collecting performance regression report at each staging release. From there, we would learn whether:

- [ ] the report generation can be run on distinct machines, operating systems
- [ ] reports looks consistent across versions, machines, operating systems
- [x] the default duration of 10 minutes is a good trade-off to quickly spot potential regressions
- [x] the current scenario provide the best value
- [x] it would be valuable to reject a release with -5% regression detected

No file, test, or entry point is named. Start by locating the existing performance-report generation and staging-release workflow, then review the checklist in this issue. Done means reports are collected for each staging release and the remaining questions about portability, consistency, duration, scenarios, and release rejection are answered.

Written by the indexing model from the issue text.