Optimize quasi-static pushing cost function's Jacobians

Current implementation uses a few intermediate matrix multiplication that can be removed and the final results just hardcoded using indexing or basic torch operations.

No files or tests are named. Locate the quasi-static pushing cost function and its Jacobian implementation, then inspect the intermediate matrix multiplications and existing torch operations before measuring or validating equivalent Jacobian results. Done means the redundant multiplications are removed and the final results remain unchanged.

Written by the indexing model from the issue text.