[XPU] Support fused_rms_norm on XPU device

[ys950902]

For cuda fused_rms_norm is set by default, support fused_rms_norm on XPU device.

[rogerxfeng8]

@tjruwase would you help review?

[polisettyvarma]

@rogerxfeng8 @tjruwase should ipex import under try except block, right ?

[tjruwase]

@polisettyvarma, good catch. I think ipex import could also be under if get_accelerator().device_name() == 'xpu': similar to cuda case.

@rogerxfeng8, can you please help fix. Thanks!

[ys950902]

@polisettyvarma, good catch. I think ipex import could also be under if get_accelerator().device_name() == 'xpu': similar to cuda case.

@rogerxfeng8, can you please help fix. Thanks!

Thanks for your suggestions, I have modified it on #436.