Using Offline Reinforcement Learning To Trial Online Platform Interventions