RoboPoint: A Vision-Language Model for Spatial Affordance Prediction for Robotics