support blockwise fp8 matmul kernel

[yizhang2077]

Motivation

support fp8 blockwise kernel (Currently only supports scale_a block shapes of 1x128 and scale_b block shapes of 128x128 for deepseek v3), mainly from vllm
correctness
python3 tests/test_fp8_blockwise_gemm.py
benchmark
python3 benchmark/bench_fp8_blockwise_gemm.py --models meta-llama/Llama-3.1-8B-Instruct
TODO (update benchmark results)

Modifications

Checklist

• Format your code according to the Code Formatting with Pre-Commit.
• Add unit tests as outlined in the Running Unit Tests.
• Update documentation / docstrings / example tutorials as needed, according to Writing Documentation.
• Provide throughput / latency benchmark results and accuracy evaluation results as needed, according to Benchmark and Profiling and Accuracy Results.

[zhyncs]

@yizhang2077 Delete all of these, we should use the version directly from 3rdparty CUTLASS instead.
BTW @BBuf will integrate a higher-performance version than this baseline, this PR won't be used directly.

[zhyncs]

I think DeepSeek V3 is sufficient, other models use per-tensor FP8, so we don't need a benchmark in that form.