Probabilistic Speech-Driven 3D Facial Motion Synthesis: New Benchmarks, Methods, and Applications