Demonstration-guided Optimal Control for Long-term Non-prehensile Planar Manipulation