Precompute multiplicative inverse when possible in op_div (#5209)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/5209

Division is generally slower than multiplication in hardware.
ghstack-source-id: 242226478

Reviewed By: kimishpatel

Differential Revision: D62412539

fbshipit-source-id: 78c659d84579c42da3dc44a8d8c88e31ea369430