[Bugfix] Restore support for larger block sizes

[kzawora-intel]

This PR reverts #10938, expands description of block size and adds assertion for CUDA-supported block sizes in CacheConfig.
While GPU kernels might not support block sizes greater than 32, other accelerators do. On HPU, going below block size 128 is very detrimental to performance, and 128 is used there by default.

[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.

🚀

[mgoin]

This is reasonable and thanks for the clear error for CUDA

[simon-mo]

Thanks for catching this

[comaniac]

Found this PR because I got an error saying block size can only be up to 32 on CUDA. However, I was using larger block sizes with FlashAttention without any issue. Just want to confirm that do we really have this constraint for all use cases on CUDA?