Can you explain the concept of neural network pruning and compression and its importance in deploying computer vision models on edge devices?