Tailoring Self-Rationalizers with Multi-Reward Distillation