How would you design a full production-grade Rust async runtime with executor, reactor, and scheduler separation?

A full runtime splits into executor (task scheduling), reactor (I/O events), and scheduler (thread balancing).

The executor manages ready tasks via queues, the reactor interfaces with OS event systems (epoll/kqueue/IOCP), and the scheduler distributes tasks across worker threads. Waker bridges reactor → executor communication. This separation ensures scalability and avoids blocking.