[nvJPEG] Integrate nvJPEG decoder (#1288)

* nvjpeg cmake

* add common decoder, nvjpeg decoder and add image name predict api

* ppclas support nvjpeg decoder

* remove useless comments

* image decoder support opencv

* nvjpeg decode fallback to opencv

* fdtensor add nbytes_allocated

* single image decode api

* fix bug

* add pybind

* ignore nvjpeg on jetson

* fix cmake in

* predict on fdmat

* remove image names predict api, add image decoder tutorial

* Update __init__.py

* fix pybind

fastdeploy/core/fd_tensor.h

@@ -54,6 +54,11 @@ struct FASTDEPLOY_DECL FDTensor {
   // other devices' data
   std::vector<int8_t> temporary_cpu_buffer;
 
+  // The number of bytes allocated so far.
+  // When resizing GPU memory, we will free and realloc the memory only if the
+  // required size is larger than this value.
+  size_t nbytes_allocated = 0;
+
   // Get data buffer pointer
   void* MutableData();