Do Large Language Models with Reasoning and Acting Meet the Needs of Task-Oriented Dialogue?