Enhancing Financial Time-Series Forecasting with Retrieval-Augmented Large Language Models