[Kernel] Use out in flash_attn_varlen_func

[WoosukKwon]

This PR uses the in-place out argument of flash_attn_varlen_func to avoid redundant copy. This is possible by the change in vllm-project/flash-attention#32

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[WoosukKwon]

Just for a note: This PR reduces the small-input latency of opt-125m from 194 ms to 184 ms

Avg latency: 0.18358359679890177 seconds
10% percentile latency: 0.18246181161375716 seconds
25% percentile latency: 0.18272388676996343 seconds
50% percentile latency: 0.18310453748563305 seconds
75% percentile latency: 0.18387063493719324 seconds
90% percentile latency: 0.18499673358164728 seconds
99% percentile latency: 0.18895185411558488 seconds