[XPU] Add flash_attn2 support for XPU (#41956)
* Add flash_attention_2 and kernels-community/flash-attn support for XPU
* Add flash-attn-2 support for XPU
* Delete deterministic algorithm for xpu
* Fix code style
* Modify repo_id to match the latest kernels-community/flash-attn2
* Fix code style
* Update
* Make quality
* Use kernels loading
* Update
* Delete invalid import
* Update comment
---------
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>