[Feature] Integrate CUTLASS FP8 GEMM into sgl-kernel

[zhyncs]

Checklist

• 1. If the issue you raised is not a feature but a question, please raise a discussion at https://github.com/sgl-project/sglang/discussions/new/choose Otherwise, it will be closed.
• 2. Please use English, otherwise it will be closed.

Motivation

ref
https://github.com/NVIDIA/cutlass/pull/1932/files

Related resources

No response

[zhyncs]

note: If the official merge does not occur by the end of the month, we will compile and use based on the sgl-project/cutlass f8_blockwise_scaling_pr_branch branch.
ref https://github.com/sgl-project/cutlass/tree/f8_blockwise_scaling_pr_branch

[HaiShaw]

@zhyncs this suppose to be NV specific.

[zhyncs]

@HaiShaw Yeah we will use it for some model on NVIDIA H100 and H200