Start local backend in cluster client mode

- Workers need a local backend to connect to even in client mode
- Add backend startup and readiness check for cluster clients
- Ensure proper cleanup on exit

vidai.py

@@ -360,9 +360,52 @@ Examples:
         print(f"Connecting to cluster master at {args.cluster_host}:{args.cluster_port}")
         print("Press Ctrl+C to stop")
 
+        # Start backend process (needed for workers to connect to)
+        backend_cmd = [sys.executable, '-m', 'vidai.backend']
+        backend_proc = subprocess.Popen(backend_cmd)
+
+        # Wait for backend to be ready
+        import time
+        import socket
+        from vidai.compat import get_socket_path, is_unix_sockets_supported
+
+        print("Waiting for backend to be ready...")
+        max_wait = 15
+        for i in range(max_wait):
+            try:
+                if is_unix_sockets_supported():
+                    # Check Unix socket
+                    sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
+                    sock.settimeout(1)
+                    sock.connect(get_socket_path('worker'))
+                    sock.close()
+                    print("Backend Unix socket is ready!")
+                    break
+                else:
+                    # Check TCP port
+                    sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+                    sock.settimeout(1)
+                    sock.connect((args.backend_host, args.backend_worker_port))
+                    sock.close()
+                    print("Backend TCP port is ready!")
+                    break
+            except (FileNotFoundError, ConnectionRefusedError, OSError):
+                if i < max_wait - 1:
+                    print(f"Waiting for backend... ({i+1}/{max_wait})")
+                    time.sleep(1)
+                else:
+                    print("Backend failed to start within timeout")
+                    backend_proc.terminate()
+                    backend_proc.wait()
+                    sys.exit(1)
+
         # Start cluster client process
         from vidai.cluster_client import start_cluster_client
+        try:
             start_cluster_client(args.cluster_host, args.cluster_port, args.token, args.optimize, args.flash, args.weight, args.cluster_shared_dir)
+        finally:
+            backend_proc.terminate()
+            backend_proc.wait()
     else:
         print("Starting Video AI Analysis Tool...")
         print(f"Server will be available at http://{args.host}:{args.port}")