CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents | Digital Library | PAMCET | PAMCET