Optimize the performance of the Combine Ant Moving function and the use of HCCL buffer#314

Yael-X merged 4 commits intosgl-project:mainsgl-project/sgl-kernel-npu:mainfrom

oagniqgnat:combine_long_seq_optimizeoagniqgnat/sgl-kernel-npu:combine_long_seq_optimizeCopy head branch name to clipboard