UniVIE: A Unified Label Space Approach to Visual Information Extraction from Form-like Documents