switch llama-swap to unified-vulkan image with FLUX.1-dev image generation

- Update deployment to unified-vulkan-2026-05-19 (includes llama-server,
  sd-server, whisper-server in one image)
- Fix binary paths: /app/llama-server -> llama-server (now on PATH)
- Migrate groups -> matrix to allow FLUX to evict the always-on 0.8B model
  when image generation is requested
- Add FLUX.1-dev Q4_K_S model via sd-server
- Configure OpenWebUI image generation to use llama-swap sd-server
- Update renovate versioning regex to treat all unified-vulkan date tags as
  patch updates for automerge

renovate.json

@@ -15,21 +15,13 @@
     {
       "matchDatasources": ["docker"],
       "matchPackageNames": ["ghcr.io/mostlygeek/llama-swap"],
-      "versioning": "regex:^v(?<major>[0-9]+)-vulkan-b(?<patch>[0-9]+)$"
+      "versioning": "regex:^unified-vulkan-(?<patch>[0-9]{4}[0-9]{2}[0-9]{2})$"
     },
     {
       "matchUpdateTypes": ["patch"],
       "automerge": true,
       "automergeType": "pr",
       "platformAutomerge": true
-    },
-    {
-      "matchDatasources": ["docker"],
-      "matchPackageNames": ["ghcr.io/mostlygeek/llama-swap"],
-      "matchUpdateTypes": ["minor"],
-      "automerge": true,
-      "automergeType": "pr",
-      "platformAutomerge": true
     }
   ]
 }