Causal Transformer for Fusion and Pose Estimation in Deep Visual Inertial Odometry