General Scene Adaptation for Vision-and-Language Navigation