A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement