nvjpegEncodeImage() not work asynchronously on latest CUDA 11.8

Continuing the discussion from nvjpegEncodeImage() not work asynchronously:

@Yuki_Ni
I wrote a test of nvjpeg encoder and it seems like there are still pagable H2D and D2H cudaMemcpyAsync in nvjpegEncodeImage() as shown below.

PS: My test environment is CUDA 11.8.