[Vulkan] Fix quantized cpu to vulkan broken by padding (#97372)
Summary:
Previous diff D43068669 introduced channel padding, and in doing so, it broke the quantized copy of cpu to vulkan tensors.
This diff updates the quantized nchw to image shaders, in order to work with padded channels.
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
```
Differential Revision: D44309956
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97372
Approved by: https://github.com/SS-JIA
