On the Re-Solving Heuristic for (Binary) Contextual Bandits with Knapsacks