Scaling Cross-Embodied Learning: One Policy for Manipulation, Navigation, Locomotion and Aviation