When it comes to WebRTC multiparty video, the most common architecture out there is SFU. The reason for it is cost and complexity. What makes it a better choice than mesh or an MCU?