Evaluating large language models' ability to understand metaphor and sarcasm using a screening test for Asperger syndrome