Filling some Int4 tensor feature gaps after tensor subclass migration

[jerryzh168]

Add AWQ Support for int4 dtypes

Similar to #2753 we want to add AWQ support for other types of int4 tensors, right now only added for Int4Tensor.

• Int4TilePackedTo4dTensor
• Int4MarlinSparseTensor
• Int4PlainInt32Tensor Support PLAIN_INT32 for AWQ on Intel GPU #3019
• Int4OpaqueTensor Support Int4OpaqueTensor for AWQ #2997

Add HQQ option

Example: #2912

• Int4OpaqueTensor
• Int4PlainInt32Tensor

[jerryzh168]

cc @Xia-Weiwen @cyxlily please feel free to take some of these since many of them are related to CPU/XPU

[Xia-Weiwen]

Sure. We will take a look. CC @liangan1 for the XPU part

[liangan1]

@jerryzh168 @Xia-Weiwen Thanks for your reminder. We will first work on the AWQ support for Int4PlainInt32Tensor. HQQ depends on the float zp support, which is not implemented by the oneDNN kernel yet. CC @xiaowangintel

[liangan1]

AWQ support for Int4PlainInt32Tensor: #3019