[Misc] Support attention logits soft-capping with flash-attn

[WoosukKwon]

This PR adds support for attention logits soft-capping in the FlashAttention backend.

This is done by

1. Moving logits_soft_cap from AttentionMetadata to AttentionImpl.
2. Using vllm-flash-attn == 2.6.1 which added support for soft-capping.

Note that vllm-flash-attn v2.6.1 uses PyTorch 2.4.0, so the PR must be merged after #6951

This will hopefully resolve many of the issues in running Gemma2 models.

[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 consists a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of default ones by unblocking the steps in your fast-check build on Buildkite UI.

Once the PR is approved and ready to go, please make sure to run full CI as it is required to merge (or just use auto-merge).

To run full CI, you can do one of these:

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

🚀

[comaniac]

LGTM