Multi-View Masked World Models for Visual Robotic Manipulation