[Vulkan] Implement Layernorm operator (#80980)
Implemented LayerNorm operator for the Vulkan backend.
Special case implementation:
- Input tensor must be 3-dim or 4-dim.
- If input tensor is 4-dim, then batch dim must be 1.
- `normalized_shape` must have length 3, i.e. [C, H, W]
- It expects weight and bias tensor arguments.
References
- PyTorch Docs > torch.nn > [LayerNorm](https://pytorch.org/docs/stable/generated/torch.nn.LayerNorm.html)
Differential Revision: [D37407311](https://our.internmc.facebook.com/intern/diff/D37407311/)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/80980
Approved by: https://github.com/SS-JIA
