[MPS] Allow `float16` input to float32 `LayerNorm` (#96430)
Only for forward pass
Subset of https://github.com/pytorch/pytorch/pull/96208
Create constant with scalar using `input_mps_dtype` and use
`reciprocalWithTensor` instead of `divisionWithPrimaryTensor:1.0
secondaryTensor:`
Fixes https://github.com/pytorch/pytorch/issues/96113
Pull Request resolved: https://github.com/pytorch/pytorch/pull/96430
Approved by: https://github.com/kulinseth
