Multi-Agent VQA: Exploring Multi-Agent Foundation Models in Zero-Shot Visual Question Answering