feat(compute): Worker-Selbstanmeldung ueber RVS + Flotten-Anzeige (Stage 2)

Jeder GPU-Dienst meldet sich beim Connect mit worker_hello {instanceId,
service, node, gpus, model} und haelt die Registry per periodischem
worker_ping {instanceId, busy} (~10s) frisch. So weiss ARIA, was wo laeuft.

- xtts/{voxtral,whisper,f5tts}/bridge.py + llm-adapter/adapter.py:
  INSTANCE_ID=service@NODE_NAME, _worker_register()-Coroutine (hello + ping),
  busy-Quelle je Worker (aktive STT-Sessions / TTS-Render / in-flight LLM);
  Task sauber gecancelt bei Reconnect.
- bridge/aria_bridge.py: self._workers-Registry + Handler worker_hello/
  worker_ping (spiegelt sat_hello), _worker_list() (35s-Offline-TTL),
  _pick_worker() (Round-Robin freie Instanz, fuer Stage-3-Routing),
  /internal/worker-list-Endpoint.
- diagnostic/server.js: workers-Map, worker_hello/worker_ping-Tracking,
  worker_update-Broadcast + worker_list-Action + on-connect-Snapshot.
- diagnostic/index.html: "Compute-Flotte"-Panel im Satelliten-Tab — pro Node
  gruppiert, mit Dienst/Modell/GPU und frei/beschaeftigt/offline-Status.

Stage 2 von 3. Reine Sichtbarkeit, kein Routing-Verhalten geaendert.

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

xtts/llm-adapter/adapter.py

@@ -47,6 +47,15 @@ RVS_TOKEN = os.getenv("RVS_TOKEN", "").strip()
 LLAMA_URL = os.getenv("LLAMA_URL", "http://llama:8081").rstrip("/")
 LLM_MODEL = os.getenv("LLM_MODEL", "qwen3-8b")
 LLM_TIMEOUT_SEC = float(os.getenv("LLM_TIMEOUT_SEC", "60"))
+
+# ── Compute-Fleet: Worker-Identitaet & Registrierung ──────────────
+# Meldet sich bei der aria-bridge (worker_hello) + periodischer worker_ping.
+NODE_NAME = os.getenv("NODE_NAME", "node").strip() or "node"
+GPU_IDS = os.getenv("NVIDIA_VISIBLE_DEVICES", "").strip()
+WORKER_SERVICE = "llm"
+INSTANCE_ID = f"{WORKER_SERVICE}@{NODE_NAME}"
+WORKER_PING_INTERVAL_S = int(os.getenv("WORKER_PING_INTERVAL_S", "10"))
+_inflight = 0   # laufende llm_requests (busy-Report im ping)
 # Qwen3 hat Thinking-Mode default AN — dann verbraet es Tokens in einem
 # <think>-Block und liefert (bei kleinem max_tokens) leeren/abgeschnittenen
 # content, ausserdem 3x langsamer. ARIAs schnelles Tier will KEIN Grübeln
@@ -124,7 +133,34 @@ async def _emit_llm_status(ws, state: str, model: str, **extra) -> None:
                 {"service": "llm", "state": state, "model": model, **extra})
 
 
+async def _worker_register(ws) -> None:
+    """Meldet diesen Worker bei der aria-bridge an (worker_hello) und haelt die
+    Flotten-Registry per periodischem worker_ping (mit busy-Status) frisch."""
+    try:
+        await _send(ws, "worker_hello", {
+            "instanceId": INSTANCE_ID, "service": WORKER_SERVICE,
+            "node": NODE_NAME, "gpus": GPU_IDS, "model": LLM_MODEL,
+        })
+        while True:
+            await asyncio.sleep(WORKER_PING_INTERVAL_S)
+            await _send(ws, "worker_ping",
+                        {"instanceId": INSTANCE_ID, "busy": _inflight > 0})
+    except asyncio.CancelledError:
+        raise
+    except Exception:
+        return   # Socket tot → still beenden; _run reconnectet + startet neu
+
+
 async def _handle_llm_request(ws, payload: dict) -> None:
+    global _last_model, _inflight
+    _inflight += 1
+    try:
+        await _do_llm_request(ws, payload)
+    finally:
+        _inflight -= 1
+
+
+async def _do_llm_request(ws, payload: dict) -> None:
     global _last_model
     req_id = payload.get("requestId", "")
     messages = payload.get("messages") or []
@@ -201,6 +237,7 @@ async def _run() -> None:
                 logger.info("RVS verbunden — llm-adapter online")
                 retry_s = 2
                 tls_fallback_tried = False
+                ping_task = asyncio.create_task(_worker_register(ws))
                 async for raw in ws:
                     try:
                         msg = json.loads(raw)
@@ -214,6 +251,10 @@ async def _run() -> None:
                     asyncio.create_task(_handle_llm_request(ws, payload))
         except Exception as e:
             logger.warning("RVS-Verbindung verloren/fehlgeschlagen: %s", e)
+            try:
+                ping_task.cancel()
+            except NameError:
+                pass
             if use_tls and RVS_TLS_FALLBACK and not tls_fallback_tried:
                 tls_fallback_tried = True
                 use_tls = False