release: 1.3.6 — subscription refresh stability + geo update hardening

Burn-in on a 1256-node Happ-bundle subscription surfaced four
refresh-related regressions: the active node disappeared after every
refresh, NodeCircle members went dangling, the same logical server
accumulated as ~22x legacy duplicates, and a double-click on
'Update Geo' raced two downloads. This release fixes all four
end-to-end.

Highlights:
- Stable-fingerprint upsert preserves Node.id across refreshes
- Per-subscription refresh mutex (409 on concurrent calls)
- NodeCircle.node_ids and active_node_id transparently remap through
  fingerprint dedup + legacy-duplicate collapse
- Geo update: per-call unique tmp path + frontend/backend mutex pair
- Boot-time active_node_id integrity check now actually runs

7 commits since v1.3.5. 23 new backend tests; full suite 608 passed.
Frontend build + tests clean.

Safe to upgrade from any 1.3.x. Next successful refresh after upgrade
auto-collapses any accumulated legacy duplicates.

Author: DaveBugg

backend/app/config.py

@@ -5,7 +5,7 @@
 # OpenAPI metadata, `/health` response, and `/system/status` so the
 # frontend can display it next to the xray version. Bump this on each
 # release — frontend keeps its own version in `frontend/package.json`.
-APP_VERSION = "1.3.5"
+APP_VERSION = "1.3.6"
 
 
 class Settings(BaseSettings):

frontend/package-lock.json

@@ -1,7 +1,7 @@
 {
   "name": "pitun-frontend",
   "private": true,
-  "version": "1.3.5",
+  "version": "1.3.6",
   "license": "BSD-3-Clause",
   "type": "module",
   "scripts": {