INST-3 · Canonical timestamps + reviewTask ↔ claimReview link

> Item derived from the fact-checking cycle analysis (**2026-07-03**). Production numbers (namespace `main`), unless noted.

### Finding (evidence)
The `published` event is **not** the public publication — only 231 of 443 tasks with history have a claim review, and the internal state precedes the claim review by **~12 median days**. The t0 is also ambiguous (claim and task are born together).

### Action
Explicit, never-overwritten fields:
- `claimSubmittedAt`
- `publishedAt` (= creation of the public claim review)
- `claimReviewId` (reviewTask ↔ claimReview link)

### Impact
Single source of truth for "published"; closes the ~12 d gap that is invisible today.

---
- **Priority:** 🟡 Medium · **Effort:** S/M
- **Dependencies:** wave 3, after INST-2/INST-4 are materialized.
- **Epic:** #2489