End-to-end Information Extraction from Archival Records with Multimodal Large Language Models