Optimize OpenCV code to use CUDA implementations

[w29ahmed]

We should take advantage of GPU optimizations since we are using a jetson, and also most of us have CUDA installed on our machines.

OpenCV has cuda implementations for a majority of the computationally expensive functions, a lot of which we are using in our computer vision node. We should start porting these over and always look for a cuda implementation first to see if it's available.

https://docs.opencv.org/4.1.0/d1/d1a/namespacecv_1_1cuda.html

This should apply to all aspects of the computer vision node: endline, traffic light, traffic sign, and lane detection