Visuo-Tactile Predictive Coding and Active Inference for Grasp and Manipulation