Generalizable Task Representation Learning for Offline Meta-Reinforcement Learning with Data Limitations