Add a xpu int8_gemm backend kernel for comfy-kitchen

[matalama80td3l]

Recently ComfyUI add int8 quant models support using the comfy-kitchen.
And please don't forget tuning on LNL :(

[matalama80td3l]

@xiangyuT