##article.return## Provably Optimal Reinforcement Learning under Safety Filtering Download Download PDF