Uncertainty-Aware Explainable Recommendation with Large Language Models