We value iteration, collaboration, and feedback. But when it comes to interviews, we still expect people to design massive systems in one go. What are we really testing, and how do we make it better?