Better to Ask in English: Cross-Lingual Evaluation of Large Language Models for Healthcare Queries