Are LLMs Good Annotators for Discourse-level Event Relation Extraction?