add SP support for `_flash_3_varlen_hub` backend (#13809)
* add `_flash_3_varlen_hub` support for SP
* make signiture clear
* update
* add accuracy test
* fix dtype
* loosen threshold because of bfloat16
* fix the break change of https://github.com/huggingface/kernels-community/commit/1471f94
* fix stale error message
* fix merge
* revert change
---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>