fix(demo): never leak the release-server IP in the public demo

- Companion QR overlay: demo builds encode the demo's own origin
  (/packages/archipelago-companion.apk ships in the web image) instead of
  the vps2 raw URL.
- Dockerfile.web/.backend: build-time scrub replaces every remaining
  occurrence of the release-server address with registry.demo.internal and
  fails the build if any survives.
- Mock backend update-mirror list, sideload help text, and changelog prose
  no longer name the server address.
- Copy demo-images workflow into .gitea/workflows/ — Gitea ignores
  .github/workflows when .gitea/workflows exists, so the CI never ran.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

neode-ui/src/components/CompanionIntroOverlay.vue

@@ -77,12 +77,17 @@
 <script setup lang="ts">
 import { ref, onMounted, watch } from 'vue'
 import * as QRCode from 'qrcode'
+import { IS_DEMO } from '@/composables/useDemoIntro'
 
 const STORAGE_KEY = 'neode_companion_intro_seen'
 // Absolute URL so the QR works when scanned by a phone (a relative path has no
 // host to resolve). Points at the companion APK hosted on the 146 release server
 // (publicly reachable) rather than the local node's /packages copy.
-const DEFAULT_DOWNLOAD_URL = 'http://146.59.87.168:3000/lfg2025/archy/raw/branch/main/neode-ui/public/packages/archipelago-companion.apk'
+// The demo serves the APK from its own public origin instead, so the QR never
+// exposes the release-server address.
+const DEFAULT_DOWNLOAD_URL = IS_DEMO
+  ? `${window.location.origin}/packages/archipelago-companion.apk`
+  : 'http://146.59.87.168:3000/lfg2025/archy/raw/branch/main/neode-ui/public/packages/archipelago-companion.apk'
 
 const visible = ref(false)
 const qrDataUrl = ref('')

neode-ui/src/views/Apps.vue

@@ -348,7 +348,7 @@
 
           <div class="mt-5 rounded-xl border border-white/10 bg-white/[0.04] p-4 text-sm text-white/65">
             <p class="font-medium text-white/80 mb-2">Easy sources</p>
-            <p>Use images from Docker Hub, GHCR, the VPS2 Gitea registry (146.59.87.168:3000), or localhost. Good first candidates: Excalidraw, Stirling PDF, FreshRSS, Wallabag, HedgeDoc, CyberChef, Mealie, or PairDrop.</p>
+            <p>Use images from Docker Hub, GHCR, the Archipelago app registry, or localhost. Good first candidates: Excalidraw, Stirling PDF, FreshRSS, Wallabag, HedgeDoc, CyberChef, Mealie, or PairDrop.</p>
           </div>
 
           <div class="mt-5 flex gap-3">

neode-ui/src/views/settings/AccountInfoSection.vue

@@ -1068,7 +1068,7 @@ init()
                   <p>Installing, updating, and removing apps no longer freezes the UI. The backend now spawns the actual work in the background and returns immediately, so the progress bar starts moving right away instead of the whole page locking up for 30+ seconds while podman pulls an image.</p>
                   <p>Install progress bar actually reflects reality now. It previously stayed at 0% until the very end because podman doesn't emit parseable progress when run without a TTY. Replaced byte-counting with seven clearly-labelled phases — Preparing, Pulling image, Creating container, Starting, Waiting for health, Finalizing, Done — each mapped to a fixed percentage so the bar only moves forward.</p>
                   <p>Launch button now appears the moment an install finishes, instead of waiting up to 60 seconds for the next container scan. After a successful install or update, the backend kicks the scanner and waits for a fresh manifest to land before flipping the app to Running, so the UI always has real port and UI-route info by the time the card becomes clickable.</p>
-                  <p>Retired the decommissioned .23 Hetzner VPS mirror. New nodes default to OVH (146.59.87.168) as Server 1 and tx1138 as Server 2 for both system updates and the app registry. Existing nodes auto-purge any saved .23 entries on next load so they stop paying connection-timeout penalties against a dead host.</p>
+                  <p>Retired the decommissioned Hetzner VPS mirror. New nodes default to the OVH mirror as Server 1 and tx1138 as Server 2 for both system updates and the app registry. Existing nodes auto-purge any saved entries for the dead mirror on next load so they stop paying connection-timeout penalties against a dead host.</p>
                   <p>Update-available badges and version comparisons work again across every app. The backend was looking for its pinned-image catalog at the wrong path and silently getting an empty result on deployed nodes, which meant the UI never showed "update available" even when a newer image was ready. The search path now matches where the image recipe actually installs the file.</p>
                   <p>Nodes with a 2 TB data drive are no longer silently configured as pruned Bitcoin nodes. The disk-size check that decides whether to enable pruning was measuring the tiny OS partition instead of the large encrypted data partition, so every archy install with a separate data volume was flipping into prune=550 mode on reconcile and deleting its historical blocks on the next bitcoin-knots restart. The check now measures the actual data partition, so full-archive nodes stay full-archive.</p>
                   <p>Recovery from a failed update no longer leaves a container permanently missing. When an app update failed partway through, the rollback path tried to restart the old container by name even though the forward path had already deleted it, leaving a hole in the node that required manual intervention. The reconcile tool now supports a --create-missing flag that rebuilds any registered container from its canonical spec, giving the update flow a safe recovery path.</p>
@@ -1226,7 +1226,7 @@ init()
                 <div class="space-y-3 text-sm text-white/80 pl-3 border-l border-white/10">
                   <p>App installs now show a real download progress bar — same accuracy as the system update bar. You'll see "Downloading: 50.5 / 200.0 MB (25%)" with a live percentage instead of a generic spinner. The bar keeps streaming even when the install falls back from one registry to another, so you'll never see a "stuck at 0%" again.</p>
                   <p>Uninstalls now show what's actually happening: "Stopping containers (2/5)", "Cleaning up volumes", "Removing app data" — labelled per app so you can fire off multiple uninstalls in parallel and watch each one's stage on its own card.</p>
-                  <p>OVH (146.59.87.168) is now baked in as Server 3 by default for both updates and the app registry — extra mirror, completely independent network path so a single-provider outage can't take everything down.</p>
+                  <p>The OVH mirror is now baked in as Server 3 by default for both updates and the app registry — extra mirror, completely independent network path so a single-provider outage can't take everything down.</p>
                 </div>
               </div>
               <!-- v1.7.29-alpha -->