Uncertainty Distillation: Teaching Language Models to Express Semantic Confidence