Beyond Egocentric Limits: Multi-View Depth-Based Learning for Robust Quadrupedal Locomotion