[MPS] Fix the crash in nan_to_num() with Float16 data type (#94220)
This PR will prevent a crash in `test_output_match_nan_to_num_cpu_float16`, that would otherwise happen with the upcoming updates to MPS Framework in Ventura (in API `logicalANDWithPrimaryTensor()`). The fix is backwards compatible with Monterey too.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94220
Approved by: https://github.com/malfet
