Use flash-attn 2.5.8 with no build isolation in docker images

[snarayan21]

What does this PR do?

Uses the latest flash attention version and installs with no build isolation (as they recommend) in docker images, rather than building from source.

What issue(s) does this change relate to?

Before submitting

• Have you read the contributor guidelines?
• Is this change a documentation change or typo fix? If so, skip the rest of this checklist.
• Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• Did you update any related docs and document your change?
• Did you update any related tests and add any new tests related to your change? (see testing)
• Did you run the tests locally to make sure they pass?
• Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[mvpatel2000]

Screenshot of launching image and importing flash attn with no error before approval 🙏

[snarayan21]

I don't think the failing GPU tests are relevant, seeing them on other PRs too. All images built successfully -- here's a screenshot of successful import with the torch 2.3.0 image:

And the flash-attn version is 2.5.8: