Add: backward pass for Pop operator

Added the backward pass implementation for the Pop operator. The backward pass stacks gradients in the correct positions of the input tensor based on the pop sequence.

Key changes:

• Implemented Pop_OpImpl::backward() method to handle gradient propagation
• Added BackwardStep counter attribute to track backward pass position
• Extended unit tests to verify gradient computation