Phase 2: Volume Wizard, GPU & device passthrough (0.2.0)

Backend:
- gpu_service: detect NVIDIA (nvidia-smi) + AMD/Intel (/dev/dri, sysfs);
  inject helpers (nvidia deploy.reservations, /dev/dri + groups + LIBVA)
- volume_service: list/orphaned/prune volumes; NFS/SMB/named/bind/tmpfs
  YAML generation (generate-yaml)
- device_service: USB/TTY/DRI detection + sandboxed host path browser
- compose_edit_service: server-side merge of volume/gpu/device fragments
- routers: volumes (+host paths), editor (services/add-volume/set-gpu/
  add-device/remove-device/set-privileged), system gpus+devices
- compose: bind-mount /dev:ro for detection

Frontend:
- split-pane StackEditor with helper panel (service picker + tabs)
- VolumeWizard (bind/named/nfs/smb/tmpfs) + HostPathBrowser
- GPUSelector, DevicePanel; api clients for volumes/editor/system

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/routers/system.py

@@ -10,6 +10,7 @@ from auth import get_current_user
 from config import settings
 from docker_client import DockerError, get_client, safe_call
 from models.user import User
+from services import device_service, gpu_service
 
 router = APIRouter(prefix="/api/system", tags=["system"])
 
@@ -88,5 +89,17 @@ def system_info(_user: User = Depends(get_current_user)) -> dict:
         "uptime_seconds": _uptime(),
         "containers_running": containers_running,
         "containers_total": containers_total,
-        "gpus": [],  # populated in Phase 2
+        "gpus": [g.to_dict() for g in gpu_service.detect_gpus()],
     }
+
+
+@router.get("/gpus")
+def gpus(_user: User = Depends(get_current_user)) -> list[dict]:
+    """Re-detect GPUs live."""
+    return [g.to_dict() for g in gpu_service.detect_gpus()]
+
+
+@router.get("/devices")
+def devices(_user: User = Depends(get_current_user)) -> dict:
+    """List host USB / serial / DRI devices for passthrough."""
+    return device_service.detect_devices()