Created by: wozna

This PR presents:

• requantize-op squash , which removes requantize operator if it occurs before the operator in which we can skip input scaling.

The image below shows this squash on the matmula operator.

For now it is enabled matmul, fc and conv.