Replication observability: link-down-since timestamp per link

**Fold-in from Chris Nelson's replication-monitoring wishlist (rider, 2026-07-15).** Promoted out of the #532 coverage-map comment into a tracked sub-issue of #437 (W8).

## Problem
There is no per-link **down-since** timestamp. Outage duration is currently reconstructed by archaeology — the two links found dead for **31 days** had no first-party record of *when* they went down.

## Ask
Expose a per-link `down_since` (or `last_connected_at`) timestamp so outage duration is computable directly. Complements the durable disconnect-reason-code work (#214 / harper#1635): reason answers *why*, this answers *how long*.

## Acceptance criteria
- Each link exposes a stable down-since timestamp in `cluster_status` and as a scrapable field (harper#1636); duration-of-outage requires no log correlation.

---
🤖 Filed by KrAIs on behalf of Kris.

Start by tracing the existing cluster_status output and per-link connection state, then inspect how monitoring fields become scrapable. Define whether down_since or last_connected_at is the stable field and verify that every link exposes it in both surfaces; completion should make outage duration computable without log correlation.

Written by the indexing model from the issue text.