Reinforcement Learning-based Fault-Tolerant Control for Quadrotor with Online Transformer Adaptation