Data Augmentation for Instruction Following Policies via Trajectory Segmentation