NVidia cuda OpticalFlow code (and sample) should utilize costBuffer

[erichfrazer]

latest OpenCV Contrib Issue: The cuda optical flow by nvidia has the flags to use a cost buffer, but there is no implementation for it. Plus, the base optical flow code doesn't have a method to upscale the cost buffer, but it does for the flowxy buffer. I fixed the sample code and the core code to enable this. I just need to check it in.

Files: cudaoptflow/include/opencv2/cudaoptflow.hpp cudaoptflow/samples/nvidia_optical_flow.cpp cudaoptflow/src/nvidiaOpticalFlow.cpp

[gabrielDiaz-performlab]

This is still an issue. ...at least, in the python & cuda implementation of opencv.