[https://nvbugs/5628952][fix] avoid cudaFree overlap with cuda graph#8903

chuangz0 wants to merge 1 commit intoNVIDIA:release/1.1NVIDIA/TensorRT-LLM:release/1.1from

chuangz0:fix_kv_cache_transceiverchuangz0/TensorRT-LLM:fix_kv_cache_transceiverCopy head branch name to clipboard