feat: merged detection output + parallel model execution

- Multiple models run simultaneously on each frame via ThreadPoolExecutor
- Detections merged with IoU deduplication (same class, high overlap → keep highest conf)
- Single merged_annotated.mp4 output (not per-model videos)
- Single model still uses existing run_pipeline for backward compat
- Active jobs banner on upload page (polls every 5s)
- GPU warning when selecting 3+ models
- 62/62 tests pass

app.py

@@ -255,6 +255,16 @@ def api_cancel_job(job_id):
     return jsonify({"error": "cannot cancel job in state " + job.status.name}), 400
 
 
+@app.route("/api/jobs/active")
+def api_active_jobs():
+    """Return count of running/pending jobs."""
+    active = [j for j in job_queue.list_jobs() if j.status.name in ('RUNNING', 'PENDING')]
+    return jsonify({
+        "count": len(active),
+        "jobs": [{"job_id": j.job_id, "status": j.status.name, "video": os.path.basename(j.video_path)} for j in active],
+    })
+
+
 @app.route("/api/jobs/<job_id>", methods=["DELETE"])
 def api_delete_job(job_id):
     """Delete a completed/failed/cancelled job."""