Imitation-Projected Programmatic Reinforcement Learning