Reinforcement Learning for Edit-Based Non-Autoregressive Neural Machine Translation