Text this: Assessing the proficiency of large language models on funduscopic disease knowledge