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

A production-grade Rust async runtime should separate the executor, scheduler, and reactor. Use a work-stealing executor with per-thread local queues, an OS-backed reactor (epoll/kqueue/IOCP) for I/O readiness, and a cooperative scheduler that manages fairness through task budgeting and Waker-driven wakeups.

The architecture is typically divided into three independent layers:

Executor

The executor runs async tasks by polling futures. Each worker thread owns a local queue to minimize lock contention and improve cache locality.

When a worker becomes idle, it steals tasks from another worker's queue instead of relying on a centralized queue.

Scheduler

The scheduler manages runnable tasks and decides which task executes next. It is responsible for:…