fix: clear GhostMaps outage state on unchanged-head recovery

docs/CURRENT-STATE.md

@@ -5128,6 +5128,7 @@ are omitted rather than framing the wrong part of the globe.
 - **Updates.** The repository is not ours, so there is no webhook. `server/providers/ghostMaps.js` starts with the first `/api/ghostmaps` request, so an install that never enables the layer never contacts GitHub.
   - It reads the branch head from git's smart-HTTP ref advertisement (`/info/refs?service=git-upload-pack`, about 0.5 KB, outside the REST rate limit) every `GHOSTMAPS_POLL_SECONDS` (15 s by default, 5 s minimum). Failures back off from 15 s to 5 min.
   - Only a new head costs REST calls: one recursive tree and one commit. An optional `GITHUB_TOKEN` raises the REST limit.
+  - After an upstream outage, a successful branch-head check clears the error and pushes a recovered catalog even when the commit is unchanged. Cached products stay in place; subsequent healthy unchanged checks stay silent.
   - The new catalog goes to `.gev-cache/ghostmaps/catalog.json` and to every open page over server-sent events (`/api/ghostmaps/events`, 25 s heartbeat, at most 64 subscribers). Default products are downloaded at once.
   - The layer reloads only products whose blob changed. The new edition's data source is added before the old one is removed. A product that leaves the catalog leaves the globe.
   - The row reads `LIVE` while the stream is open. A five-minute catalog read is the backstop.