VRDSynth: Synthesizing Programs for Multilingual Visually Rich Document Information Extraction