feat(gpu): GPU health check + RENDER_DEVICE_USED token + strict mode

- gpu_probe.py: Blender script that probes OPTIX/CUDA/HIP/ONEAPI and
  exits 1 on no GPU — used at startup + on-demand from Admin UI
- blender_render.py, still_render.py, turntable_render.py: emit
  RENDER_DEVICE_USED: engine=CYCLES device=GPU|CPU compute_type=...
  after GPU activation; exit 2 when CYCLES_DEVICE=gpu and CPU fallback
- render_blender.py: parse RENDER_DEVICE_USED token into render_log
  (device_used, compute_type, gpu_fallback); handle exit code 2 as
  explicit GPU strict-mode failure
- check_version.py: check_gpu() runs gpu_probe.py at container startup;
  CYCLES_DEVICE=gpu aborts startup if no GPU found
- docker-compose.yml: CYCLES_DEVICE=${CYCLES_DEVICE:-auto} env var
- gpu_tasks.py: probe_gpu Celery task on thumbnail_rendering queue;
  saves result to system_settings.gpu_probe_last_result; beat every 30min
- worker.py: POST /probe/gpu (trigger) + GET /probe/gpu/result (last result)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/app/tasks/gpu_tasks.py

@@ -0,0 +1,88 @@
+"""Celery task for GPU health probe."""
+import logging
+from app.tasks.celery_app import celery_app
+
+logger = logging.getLogger(__name__)
+
+
+@celery_app.task(name="app.tasks.gpu_tasks.probe_gpu", queue="thumbnail_rendering")
+def probe_gpu() -> dict:
+    """Run Blender GPU probe on the render-worker. Stores result in system_settings."""
+    import subprocess
+    import json
+    from datetime import datetime, timezone
+    from pathlib import Path
+    from app.services.render_blender import find_blender
+
+    result = {
+        "status": "unknown",
+        "device_type": None,
+        "devices": [],
+        "error": None,
+        "probed_at": datetime.now(timezone.utc).isoformat(),
+    }
+
+    try:
+        blender_bin = find_blender()
+        if not blender_bin:
+            result["status"] = "error"
+            result["error"] = "Blender binary not found — check BLENDER_BIN env or PATH"
+        else:
+            probe_script = Path("/render-scripts/gpu_probe.py")
+
+            if not probe_script.exists():
+                result["status"] = "error"
+                result["error"] = f"gpu_probe.py not found at {probe_script}"
+            else:
+                proc = subprocess.run(
+                    [blender_bin, "--background", "--python", str(probe_script)],
+                    capture_output=True, text=True, timeout=60,
+                )
+
+                for line in proc.stdout.splitlines():
+                    if "GPU_PROBE_OK:" in line:
+                        result["status"] = "ok"
+                        # Parse device_type and devices from line:
+                        # GPU_PROBE_OK: device_type=OPTIX devices=[...]
+                        parts = line.split("GPU_PROBE_OK:", 1)[1].strip()
+                        for p in parts.split():
+                            if p.startswith("device_type="):
+                                result["device_type"] = p.split("=", 1)[1]
+                        break
+                    elif "GPU_PROBE_FAIL:" in line:
+                        result["status"] = "failed"
+                        result["error"] = line.split("GPU_PROBE_FAIL:", 1)[1].strip()
+                        break
+
+                if result["status"] == "unknown":
+                    result["status"] = "failed" if proc.returncode != 0 else "unknown"
+                    result["error"] = proc.stderr[:500] if proc.stderr else "No probe output"
+
+    except subprocess.TimeoutExpired:
+        result["status"] = "error"
+        result["error"] = "GPU probe timed out after 60s"
+    except Exception as exc:
+        result["status"] = "error"
+        result["error"] = str(exc)
+
+    # Save to system_settings
+    _save_probe_result(result)
+    return result
+
+
+def _save_probe_result(result: dict) -> None:
+    import json
+    from sqlalchemy import create_engine, text
+    from app.config import settings as app_settings
+
+    sync_url = app_settings.database_url.replace("+asyncpg", "")
+    eng = create_engine(sync_url)
+    try:
+        with eng.connect() as conn:
+            conn.execute(text("""
+                INSERT INTO system_settings (key, value) VALUES (:key, :value)
+                ON CONFLICT (key) DO UPDATE SET value = EXCLUDED.value
+            """), {"key": "gpu_probe_last_result", "value": json.dumps(result)})
+            conn.commit()
+    finally:
+        eng.dispose()