MV-UMI: A Scalable Multi-View Interface for Cross-Embodiment Learning