Optimizing Credit Limit Adjustments Under Adversarial Goals Using Reinforcement Learning