[Vulkan] Fix broadcasting in quantized elementwise ops (#97554)
Summary: Fixes broadcasting along the channel and batch dimensions in quantized add, sub, mul and div
Test Plan:
```
buck run --target-platforms ovr_config//platform/macos:arm64-fbsource -c pt.vulkan_full_precision=1 //xplat/caffe2:pt_vulkan_quantized_api_test_binAppleMac\#macosx-arm64
```
Reviewed By: SS-JIA
Differential Revision: D44359706
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97554
Approved by: https://github.com/SS-JIA
