Minimizing Weighted Counterfactual Regret with Optimistic Online Mirror Descent