Why Pathmind Failed: Deep Reinforcement Learning for Industrial Optimization