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

A scalable runtime separates executor, reactor, and scheduler with work-stealing queues and Waker-driven task resumption.

Workers maintain local deques for cache locality. Idle workers steal tasks to balance load. The reactor integrates OS event polling (epoll/kqueue/IOCP). Waker bridges I/O completion and task rescheduling. This avoids global locks and ensures scalability under high concurrency.