Compare PageSpeed Insights before and after a deploy, and fail CI on regressions

Audit the same URLs before and after a release, diff LCP, CLS and TBT with a tolerance, and exit non-zero on regressions so CI can block them.