Evaluating Large Language Models for Health-related Queries with Presuppositions