[fix] Add 1 and draft_token_num to seq_len when overlap scheduling is enabled during memory estimation#5343

HuiGao-NV merged 6 commits intoNVIDIA:mainNVIDIA/TensorRT-LLM:mainfrom

HuiGao-NV:extra_token_for_overlapHuiGao-NV/TensorRT-LLM:extra_token_for_overlapCopy head branch name to clipboard