Nightly evals

Declares the nightly eval runner standing: a deterministic schedule that re-runs the confirmed eval cases every night, as the person who confirmed each one.

• Purpose: Every night from the declared start hour (01:00 UTC by default), re-runs up to the declared cap of confirmed eval cases (15 cases, 3 conversations each, by default), oldest last run first, each as its stamped confirmer on an explicit EU-routed model and effort, and writes each result onto its case for the weekly digest.
• Triggers: Nightly schedule — hourly check on Feedback/EvalNights; one night per day from Feedback:Evals:Nightly:StartHourUtc, when Feedback:Evals:Nightly:Enabled is true
• Runs on: an instance that enables it by configuration (off by default everywhere)
• Wired: yes
• Queue path: Feedback/EvalNights
• Doc path: Feedback/AnswerFeedback
• Note: No model decides what runs: the night is planned in code. Each night's record is `Feedback/_EvalNights/{date}`; each case's result is its `lastRun`.