Beyond the Assistant Turn: User Turn Generation as a Probe of Interaction Awareness in Language Models