See, Think, Confirm: Interactive Prompting Between Vision and Language Models for Knowledge-based Visual Reasoning