Active Inference or Control as Inference? A Unifying View