feat(oci): GPU selection per host and per image, one notification per update, and App tab for stack containers

- Immich asks what runs its video and its recognition in one menu, in both
  modes, and gives the GPU to the server and to Machine learning; AMD uses ROCm
- Frigate, Ollama, llama.cpp, Faster Whisper and Piper take the image built
  for the chosen GPU
- The acceleration menu offers only what the host can run
- An update or a recreation sends one notification with its result instead of
  the stop, backup and start of each container
- A private bridge with nothing connected is not reported as down
- Secondary containers of a stack appear in the App tab with their version and
  logo; Secure Gateway shows the same update state in both views
- A mistyped value in the wizard asks the same question again

oci/tests/test_ai_gpu_profiles.py

@@ -0,0 +1,59 @@
+"""The AI applications whose image depends on the GPU take the image and the
+devices of the profile that is chosen."""
+
+from pathlib import Path
+import json
+import sys
+import unittest
+from unittest.mock import patch
+
+ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(ROOT / "src"))
+sys.path.insert(0, str(Path(__file__).resolve().parent))
+
+from proxmenux_oci.catalog import Catalog
+from proxmenux_oci.installer import DEFAULT_MODE, build_deployment
+from test_advanced_flow_order import RecordingUI, storages
+
+RENDER, KFD = "/dev/dri/renderD128", "/dev/kfd"
+EXPECTED = {
+    "ollama": {"cpu": (":latest", []), "nvidia": (":latest", ["nvidia-runtime"]), "rocm": (":rocm", [RENDER, KFD])},
+    "llamacpp": {"cpu": (":server", []), "nvidia": (":server-cuda", ["nvidia-runtime"]),
+                 "rocm": (":server-rocm", [RENDER, KFD]), "intel": (":server-intel", [RENDER])},
+    "faster-whisper": {"cpu": (":latest", []), "nvidia": (":gpu", ["nvidia-runtime"])},
+    "piper": {"cpu": (":latest", []), "nvidia": (":gpu", ["nvidia-runtime"])},
+}
+SOURCES = {"ollama": "overlays", "llamacpp": "curated", "faster-whisper": "overlays", "piper": "overlays"}
+
+
+# Every profile is offered: what the host has is checked in its own test.
+@patch("proxmenux_oci.installer.host.gpus", return_value={"intel": ["/dev/dri/renderD128"], "amd": ["/dev/dri/renderD128"], "nvidia": True})
+@patch("proxmenux_oci.installer._profile_usable", return_value=True)
+@patch("proxmenux_oci.i18n.language", return_value="en")
+@patch("proxmenux_oci.installer.host.storages", side_effect=storages)
+@patch("proxmenux_oci.installer.host.bridges", return_value=[{"iface": "vmbr0", "cidr": "192.0.2.10/24"}])
+@patch("proxmenux_oci.installer.host.timezone", return_value="Europe/Madrid")
+class AiGpuProfileTests(unittest.TestCase):
+    catalog = Catalog(ROOT)
+
+    def test_each_profile_installs_its_image_with_its_devices(self, *_):
+        for app, profiles in EXPECTED.items():
+            hardware = self.catalog.compose(app)["proxmox"]["installer_profile"]["hardware_acceleration"]
+            self.assertEqual([profile["id"] for profile in hardware["profiles"]], list(profiles), app)
+            self.assertEqual(hardware["default"], "cpu", app)
+            for profile, (tag, devices) in profiles.items():
+                template = self.catalog.compose(app)
+                plan = build_deployment(template, RecordingUI({hardware["prompt"]: profile}), DEFAULT_MODE)
+                self.assertTrue(template["container_contract"]["image"]["reference"].endswith(tag), (app, profile))
+                self.assertEqual([device.get("host_path") or device["kind"] for device in plan["devices"]],
+                                 devices, (app, profile))
+
+    def test_the_shipped_copy_matches_its_source(self, *_):
+        read = lambda place, app: json.loads((ROOT / f"catalog/{place}/{app}.json").read_text())[
+            "proxmox"]["installer_profile"]["hardware_acceleration"]
+        for app, source in SOURCES.items():
+            self.assertEqual(read(source, app), read("apps", app), app)
+
+
+if __name__ == "__main__":
+    unittest.main()