examples: typed_actor, named_genserver, worker_pool for the new surface

typed_actor and named_genserver land as written (the target-ergonomics spec):
identity-bound Pid<A> vs durable Name<M>, and a gen_server addressed by a
durable ServerName. worker_pool is reworked from the spec into a self-draining
program: workers retire on a sentinel and report their tally, and the
dispatcher waits the pool out — so it terminates on its own rather than leaning
on root-exit teardown, while still exercising the full typed surface
(spawn_addr / join / pick_as / members_as / dispatch) alongside the untyped
escape hatch (members / pick + send_dyn).

All three build unchanged-against-spec (typed_actor, named_genserver) and run
to completion.

examples/worker_pool.rs

@@ -0,0 +1,95 @@
+//! A typed worker pool over process groups (`pg`).
+//!
+//! Workers enroll in a named group; the dispatcher reaches them through it.
+//! Because the pool is homogeneous — every member is a `Worker` — the group's
+//! typed reads (`pick_as` / `members_as`) and the `dispatch` combinator hand
+//! members back as `Pid<Worker>`, so every send is an ordinary compile-checked
+//! `send_to` rather than the untyped escape hatch. The untyped reads
+//! (`members` / `pick` + `send_dyn`) stay available for identity-only use —
+//! counting, logging, monitoring — where the message type isn't known.
+//!
+//! The pool drains itself: each worker retires on a sentinel and reports its
+//! tally back, and the dispatcher waits the pool out before returning. (A pool
+//! left running would be stopped anyway when the root actor exits, but draining
+//! explicitly keeps the example deterministic.)
+
+use smarm::{
+    channel, dispatch, join, members, members_as, pick, pick_as, run, send_dyn, send_to,
+    spawn_addr, Addressable, Pid,
+};
+
+/// The pool's worker actor. One message type, carried by `Pid<Worker>`.
+struct Worker;
+impl Addressable for Worker {
+    type Msg = Job;
+}
+
+/// A unit of work, or the sentinel that retires a worker.
+enum Job {
+    Task { id: u64 },
+    Retire,
+}
+
+const POOL: &str = "pool";
+const WORKERS: u64 = 4;
+
+fn main() {
+    run(|| {
+        // Mint four typed workers and enroll them. `spawn_addr` yields a
+        // `Pid<Worker>`; `join` takes a typed pid directly, erasing internally.
+        // Each worker reports how many tasks it handled over a shared channel,
+        // so the dispatcher can wait the pool out at the end.
+        let (done_tx, done_rx) = channel::<(u64, u64)>();
+        for id in 0..WORKERS {
+            let done_tx = done_tx.clone();
+            let w: Pid<Worker> = spawn_addr::<Worker>(move |rx| {
+                let mut handled = 0;
+                while let Ok(job) = rx.recv() {
+                    match job {
+                        Job::Task { id: job } => {
+                            println!("worker {id} handling job {job}");
+                            handled += 1;
+                        }
+                        Job::Retire => break,
+                    }
+                }
+                done_tx.send((id, handled)).unwrap();
+            });
+            join(POOL, w);
+        }
+        drop(done_tx); // from here only the workers hold senders
+
+        // Pick one live member and hand it a job — typed end to end.
+        if let Some(w) = pick_as::<Worker>(POOL) {
+            send_to(w, Job::Task { id: 1 }).unwrap();
+        }
+
+        // `dispatch` rolls pick-a-live-member-and-send into one call, returning
+        // the member it reached (or handing the job back if the pool is empty).
+        let _ = dispatch::<Worker>(POOL, Job::Task { id: 2 });
+
+        // Fan a job out to the whole pool — typed, so each send is `send_to`.
+        for w in members_as::<Worker>(POOL) {
+            let _ = send_to(w, Job::Task { id: 3 });
+        }
+
+        // The untyped reads stay available for identity-only use. To message a
+        // member reached this way, the explicit `send_dyn` escape hatch names
+        // the message type.
+        println!("pool size: {}", members(POOL).len());
+        if let Some(any) = pick(POOL) {
+            let _ = send_dyn::<Job>(any, Job::Task { id: 4 });
+        }
+
+        // Retire every worker, then drain their tallies so the program winds
+        // down on its own.
+        for w in members_as::<Worker>(POOL) {
+            let _ = send_to(w, Job::Retire);
+        }
+        for _ in 0..WORKERS {
+            if let Ok((id, handled)) = done_rx.recv() {
+                println!("worker {id} retired after {handled} jobs");
+            }
+        }
+    });
+}