Does Reinforcement Learning Incentivize Reasoning Capacity in LLMs?