[Speculative Decoding][MTP] Support static CacheKV C8 quantization and optimize memory usage#5155

freeliuzc merged 2 commits intoPaddlePaddle:developPaddlePaddle/FastDeploy:developfrom

freeliuzc:merge_mtp_c8_and_mem_optfreeliuzc/FastDeploy:merge_mtp_c8_and_mem_optCopy head branch name to clipboard