feat(admin): LLM-/GPU-Status-Karte (read-only) — Plan-Schritt 2

- app/admin_llm.py: read-only Statusabfragen ohne sudo (docker ps, ollama ps,
  nvidia-smi, systemctl --user) mit sicheren Defaults bei fehlenden Tools.
- GET /api/admin/llm/status (require_admin): Backend, Modell, Backend-Status,
  geladene Ollama-Modelle, GPU-Auslastung, Gateway-Dienst-Status.
- Admin Status-Tab: LLM-Backend-Karte mit GPU-Balken.
- Tests: Auth-Gate + Antwortschema (160 grün).
- Doku §7.5: Status-Tab um LLM/GPU-Karte ergänzt.

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

app/api/admin.py

@@ -6,6 +6,7 @@ from fastapi import APIRouter, Depends, HTTPException, Request, WebSocket, WebSo
 from fastapi.responses import FileResponse
 from pydantic import BaseModel
 
+from app.admin_llm import llm_status
 from app.auth import is_admin_user, require_admin, require_admin_or_user
 from app.config import settings
 from app.dependencies import get_store
@@ -187,6 +188,14 @@ async def export_db():
     )
 
 
+# ── LLM-/System-Status (read-only) ──────────────────────────────────────────
+
+@router.get("/admin/llm/status", dependencies=[Depends(require_admin)])
+async def get_llm_status():
+    """Read-only Status: aktives Backend, Modell, GPU-Auslastung, Dienste."""
+    return await llm_status()
+
+
 # ── Aussprache-Lexikon CRUD ─────────────────────────────────────────────────
 
 def _read_pronunciation(lang: str) -> dict: