Evaluation of Synthetic Datasets for Conversational Recommender Systems