Accelerating Max-Pool sub-graph

Achieved 53% speed-up with custom designed CUDA kernel leveraging kernel fusion, padding elimination & memory footprint reduction in in YOLOV4 (atm it was cool)

Addressed performance bottlenecks (unnecessary computational cycles, excessive memory allocation, and limited fusion opportunities) through strategic kernel launch configurations, and adaptability to accommodate larger filter chains and non-singular stride patterns

~ More description and cool demos coming soon

<!-- More description and cool demos coming soon -->