KDRL: Post-Training Reasoning LLMs via Unified Knowledge Distillation and Reinforcement Learning