On the structure of atmospheric warming in models and observations: Implications for the lapse rate feedback