TeaMs-RL: Teaching LLMs to Generate Better Instruction Datasets via Reinforcement Learning