Enhancing Audiovisual Speech Recognition through Bifocal Preference Optimization