Vision-Language Navigation With Self-Supervised Auxiliary Reasoning Tasks