feat(supervisor): one_for_all and rest_for_one strategies + ordered shutdown

Adds a Strategy enum (OneForOne default, OneForAll, RestForOne) selected
via OneForOne::strategy(). The triggering child's Restart policy still
decides whether anything restarts; the strategy decides which siblings
are cycled with it:
  - OneForAll : all live siblings
  - RestForOne: siblings started after the failed child

Group restarts stop the affected survivors cooperatively (request_stop)
in reverse start order, await each one's termination signal on the
existing funnel (no new channel, no select), then restart the whole set
in start order. Signals that arrive for a child we aren't currently
stopping are stashed and replayed by the main loop. A Stopped signal now
counts as abnormal for the restart decision.

On giving up (intensity cap) or any exit with survivors, an ordered
shutdown stops the remaining children in reverse start order instead of
leaking them; on the normal all-settled exit it's a no-op.

The struct keeps the OneForOne name for compatibility (existing tests
unchanged); rename is a later refactor.

src/lib.rs

@@ -46,7 +46,7 @@ pub use scheduler::{
     block_on_io, request_stop, run, self_pid, sleep, spawn, spawn_under, wait_readable,
     wait_writable, yield_now, JoinError, JoinHandle,
 };
-pub use supervisor::{ChildSpec, OneForOne, Restart, Signal};
+pub use supervisor::{ChildSpec, OneForOne, Restart, Signal, Strategy};
 
 // ---------------------------------------------------------------------------
 // check!()