Are LLMs Good Zero-Shot Fallacy Classifiers?