SPEC-RL: Accelerating On-Policy Reinforcement Learning via Speculative Rollouts - View it on GitHub