3D Crowd Counting via Geometric Attention-Guided Multi-view Fusion