Towards High-Consistency Embodied World Model with Multi-View Trajectory Videos