Double Entendre: Robust Audio-Based AI-Generated Lyrics Detection via Multi-View Fusion