Inside Kimi 1.5: A self-contained summary of its reinforcement learning efforts