V-JEPA 2: Self-Supervised Video Models Enable Understanding,Prediction,Planning