I guess, that depends on the complexity of individual screens. You can start with one component per screen. And then if you feel that a particular screen is bloated, you can extract things into child components. The screen component's entry point usually remains unchanged, its child components are implementation details.