Fix text_llm_runner kv cache pos count and use it for generate()#15286

kirklandsign merged 2 commits intomainpytorch/executorch:mainfrom

text-llm-pospytorch/executorch:text-llm-posCopy head branch name to clipboard