Expand description
Execution domains organized by workload shape. Long-lived actors run on the actor pool (two worker groups:
coordination for tiny high-frequency handlers, flow for heavy flow execution) with per-worker run queues and
pinned dispatch. Short-lived work (per-request actors, one-shot jobs) runs on the task pool. Data-parallel work
runs on the compute pool (rayon behind an install-only API). Async I/O runs on the embedded tokio runtime.
Native targets get the real pools; single-threaded and DST targets get the inline stub variant. The Pools type
both impls hand back is what SharedRuntime carries around.