feat(signals): scheduled follow-up checks on a report

Problem A report is a point-in-time claim. Nothing in the system re-measures whether the claim still holds or whether the fix it prompted worked. Follow-up exists today only as prose: • Every scout prompt carries a "Follow up on your own past work" section (products/signals/backend/scout_harness/prompt.py) that asks the scout to write followup:<skill>:<entity> scratchpad entries with a validate-after date, a probe, and a baseline. The section says outright that there is no schedule and no harness trigger. • signals-scout-inbox-validation is a whole canonical scout built around the same workaround. Its queue is pending: / addressed: / noise: scratchpad entries, and it validates about three reports per run. • The scratchpad is untrusted at read time, so a stored probe can never be executed as written. Every check is re-derived from the live report each run, which costs a sandbox turn per check. • The research agent has no way to say "…