Efficient Navigation in Unknown Indoor Environments with Vision-Language Models