Psychologically-informed chain-of-thought prompts for metaphor understanding in large language models