v3.7.0

Add a week-over-week panel to the Reliability page: compares each routine's last 7 days of finished runs against the 7 days before — runs, failures, success-rate change in percentage points and p50 duration change — with a WORSE / QUIET / NEW / BETTER / STEADY verdict, worst first, and flags when the fetched sample does not reach back 14 days.