[MPS] Move bitwise ops registration to native_functions.yaml (#98908)
Per the offline discussion, there is no technical reason/limitation to have to register bitwise ops using `TORCH_LIBRARY_IMPL`.
Move the registration to `native_functions.yaml` for an easier lookup and consistent registration patterns as other mps ops.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/98908
Approved by: https://github.com/kulinseth
