Voila: Voice-Language Foundation Models for Real-Time Autonomous Interaction and Voice Role-Play