Optimal Rewards versus Leaf-Evaluation Heuristics in Planning Agents