Room-Across-Room: Multilingual Vision-and-Language Navigation with Dense Spatiotemporal Grounding