Batched Contextual Reinforcement: A Task-Scaling Law for Efficient Reasoning