Exploring the Zero-Shot Potential of Large Language Models for Detecting Algorithmically Generated Domains