nx_async_rt.nx

async_rt.nx -- cooperative async task runtime. Phase A: runtime data structures + scheduler. Language-level `async fn` / `await` (EFFICIENCY_ROADMAP §4.3) is a parser change pending; this module gives callers enough machinery to hand-code stackful coroutines today + get lifted into real async/await syntax later without breaking the API. Design: single-threaded cooperative scheduler on one kernel thread. Each task = (function_pointer, state_blob). Scheduler runs ready queue to exhaustion, then epolls for IO readiness + wakes blocked tasks. Why single-threaded: thread safety without atomics. Multi- thread scheduling adds complexity that belongs in a later phase once the sequential model is proven. Used by: HTTP server accept loops, DNS resolver, file watcher, anything that would otherwise thread-per-connection. Invariants: AR1 Scheduler `run_until_idle` returns when all tasks are blocked waiting for IO + nothing's ready. Caller does the epoll_wait before re-entering. AR2 `task_spawn` always succeeds if slot cap not hit; returns task id. AR3 `task_yield` marks the calling task ready + saves its state via the caller-supplied resume hook.