[MPS] Fixes in backward functions of the MPS ops (#95327)
* [MPS] Fix bilinear backward pass (#94892)
Fixes backward pass for bilinear.
Summary of changes:
- bilinear op is able to produce **contiguous, non-view** tensors with a storage offset, such as: shape=`[1, 1, 1, 1]`, `storage_offset=12`. This seems a weird case, but it is valid, and for these type of tensors we wouldn't be able to gather/scatter since we look at the view flag (which is not set here). This change looks into `storage_offset` only rather than the is_view flag which is not being set
- **reduction sum** must return a zeroed out output if passing an input with 0 elements (e.g a shape of (0, 5)).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94892
Approved by: https://github.com/kulinseth
* [MPS] Fix the crash in elu_backward() (#94923)
Fixes a crash where the inputTensor could go null and cause a crash.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94923
Approved by: https://github.com/DenisVieriu97, https://github.com/kulinseth
* [MPS] Fix prelu backward pass (#94933)
Allocate the correct shape for the weights gradient
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94933
Approved by: https://github.com/razarmehr
* [MPS] Fix embedding_backward() issue with Float16 (#94950)
- Casting the float16 input tensor to float32 and cast back the output tensor
Pull Request resolved: https://github.com/pytorch/pytorch/pull/94950
Approved by: https://github.com/DenisVieriu97
---------
Co-authored-by: Denis Vieriu <dvieriu@apple.com>
Co-authored-by: Ramin Azarmehr <razarmehr@apple.com>
