Improving Optimization Bounds using Machine Learning: Decision Diagrams meet Deep Reinforcement Learning