Censoring-Aware Tree-Based Reinforcement Learning for Estimating Dynamic Treatment Regimes with Censored Outcomes