Simultaneous Object Reconstruction and Grasp Prediction using a Camera-centric Object Shell Representation