Extending Information Bottleneck Attribution to Video Sequences