Rethinking the Evaluation of Dialogue Systems: Effects of User Feedback on Crowdworkers and LLMs