Zero-shot Large Language Models for Automatic Readability Assessment