SMPLer: Taming Transformers for Monocular 3D Human Shape and Pose Estimation