orin nx 使用tensorrt推理报错

温馨提示：根据社区不完全统计，按照模板提问，可以加快回复和解决问题的速度

环境

• 【FastDeploy版本】： 最新版本

• 【编译命令】
  mkdir build && cd build
  cmake ..
  -DENABLE_ORT_BACKEND=ON
  -DENABLE_TRT_BACKEND=ON
  -DWITH_GPU=OFF
  -DBUILD_ON_JETSON=ON
  -DCMAKE_INSTALL_PREFIX=${PWD}/compiled_fastdeploy_sdk
  -DENABLE_VISION=ON
  -DCMAKE_CXX_STANDARD=17
  -DENABLE_VISION=ON
  -DOPENCV_DIRECTORY=/usr/lib/cmake/opencv4/
  make -j4 && make install

• 【系统平台】: arrch64

• 【硬件】：

• 【编译语言】： C++

问题日志及出现问题的操作流程

• 附上详细的问题日志有助于快速定位分析
  [ERROR] fastdeploy/runtime/backends/tensorrt/trt_backend.cc(239)::log 3: [runtime.cpp::~Runtime::344] Error Code 3: API Usage Error (Parameter check failed at: runtime/rt/runtime.cpp::~Runtime::344, condition: mEngineCounter.use_count() == 1. Destroying a runtime before destroying deserialized engines created by the runtime leads to undefined behavior.
  )
  在使用tensorrt推理时，只能够推理一张图片

Start with fastdeploy/runtime/backends/tensorrt/trt_backend.cc around line 239 and reproduce the reported TensorRT inference flow using the provided Jetson build configuration. Investigate the runtime and deserialized-engine lifetime reported in the log; done means multiple images can be inferred without the runtime-destruction error.