Iteration of Thought: Leveraging Inner Dialogue for Autonomous Large Language Model Reasoning