adding flash attention and xformer memory efficient through PT SDPA

[HamidShojanazeri]

What does this PR do?

This PR adds the Flash Attention and Xformer mem-efficient kernel through PT SDPA, this work has been integrated with optimum library of HF, read more about here.

Tested on 7B for FSDP only had a nice 30% speed up, for FSDP+PEFT 5% and not much on PEFT+quantization/1 gpu.

Fixes # (issue)
No related issue.

Feature/Issue validation/testing

Please describe the tests that you ran to verify your changes and relevant result summary. Provide instructions so it can be reproduced.
Please also list any relevant details for your test configuration.

• Test A : Logs/ perf number of with out this feature with 10 steps : avg epoch time 55.17s
  Logs for Test A
  

• Test B : Logs/ perf number of with this feature with 10 steps : avg epoch time 42.44
  Logs for Test B

Before submitting

• [x ] This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue? Please add a link
  to it if that's the case.
• [x ] Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Thanks for contributing 🎉!

[chauhang]

@HamidShojanazeri Thanks for adding the BT optimizations. Please see the comments inline

[chauhang]

Please fix the issues and also attach the logs for the inference speedup.