[Feature] integrate FlashInfer Blackwell kernels

[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

as titled

Related resources

No response

[zhyncs]

cc @yzh119 @tqchen

[zhyncs]

ref flashinfer-ai/flashinfer#1039

[zhyncs]

ref #5538

[Fridge003]

Todo List

• Upgrade to Flashinfer 0.2.5 Bump Flashinfer to 0.2.5 #5870 @Fridge003
• Integrate MHA kernels [Feature] Support Flashinfer fmha on Blackwell #6930 @Qiaolin-Yu @NorthmanPKU
• Integrate MLA kernels [perf][sgl-kernel] extend cutlass_mla_decode to support num_head < 128 #6929
• Integrate fp8 blockwise/grouped Gemm kernels [Feature] Support Flashinfer fp8 blockwise GEMM kernel on Blackwell #6479 Remove contiguous before Flashinfer groupwise fp8 gemm #6804 @Fridge003
• Integrate AllReduce kernels from feat: add trtllm all-reduce (non-MoE) flashinfer-ai/flashinfer#1096
• PDL Support [sgl-kernel] upgrade to flashinfer 0.2.5 to enable PDL in activaton #6722