MinerU2.5: A Decoupled Vision-Language Model for Efficient High-Resolution Document Parsing Paper โข 2509.22186 โข Published Sep 26, 2025 โข 139