Remove completion handler from custom kernel

mlx/backend/cuda/custom_kernel.cpp

@@ -374,7 +374,6 @@ void CustomKernel::eval_gpu(
         }
       });
   encoder.add_kernel_node(kernel, grid, block, shared_memory_, args.args());
-  encoder.add_completed_handler([args = std::move(args)]() {});
 }
 
 } // namespace mlx::core::fast