Add binary op support for height and width packing GPU layouts (#2516)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/2516

## Context

Enable `binary_op` to support inputs that are `HEIGHT_PACKED` and `WIDTH_PACKED`.
ghstack-source-id: 219281504
exported-using-ghexport

Reviewed By: jorgep31415

Differential Revision: D55031044

fbshipit-source-id: 0db40bbdf1ab6a1b6f38c9b67e6dcc10a431ad5f