INST-6 · workSessions (time-series) — active effort across all stages

> Item derived from the fact-checking cycle analysis (**2026-07-03, v2**). Production numbers (namespace `main`), unless noted.

### Finding (evidence)
Separating queue from active effort is only possible today in `draft`, via gaps between saves (<30 min), N=64 — the other stages emit no work signal. Result: **"78% of the clock is queue"** is an aggregate estimate, not a per-stage measurement.

### Action
A **time-series** collection of edit / save / focus events by `reviewTaskId` + `stage` + `userId`; sum into `cycleMetrics.editingActiveH`.

### Impact
"How much human time a fact-check costs" becomes a robust, per-stage number, not an estimate.

---
- **Priority:** 🟠 High-leverage · **Effort:** L (large)
- **Note:** engineering flags this as P1-by-leverage. Keep effort **L** and **sequence after INST-2 (#2492)** — it needs the stages instrumented first — but raise it above the other 🟡 items in the value queue.
- **Dependencies:** INST-2 (#2492).
- **Epic:** #2489