Active Prompt Learning in Vision Language Models