RL-Finetuned LLMs for Privacy-Preserving Synthetic Rewriting