Support StreamK when scheduling

[i-zhen]

Could DeepGEMM support StreamK schedule while K > 3N in some situation? That will make gemm faster. Like this issue: vllm-project/vllm#12978

[LyricZhao]

We may later (no sure the date) add such support, currently no plan. Thanks for your feedback, anyway :)

[Insideyyy]

We added split-k optimization in PR #186, which delivers notable speedups when m x n is too small to saturate SMs.