Eager-handoff trigger for distributed sessions.

Horde.DynamicSupervisor does NOT redistribute a node's children when that node dies (with members: :auto the dead member is removed from the CRDT, so the reclaim path never fires — see the Phase-7 design §7.6). This reaper provides selective eager handoff instead: it monitors cluster nodes and, on :nodedown, restarts the :eager sessions whose server died with the lost node.

A session is reaped when (1) its turn is in flight (SessionStore.list_resumable/1 lists the sessions whose last turn-state save was :eager and unfinished), (2) it is not registered anywhere (SessionRegistry.whereis/2 == :none), and (3) its persisted turn state is still non-terminal (this guards against an index member a crash left behind). A deleted session (Turn.Session.delete/1) is never listed. It is restarted as a thin spec under the supervisor; the started Turn.Server reconstructs config from the persisted template and Turn.resume/2s the in-flight turn — no inbound request.

The restarted server restores the plain-data options saved with the template (:approval_timeout_ms, :on_approval_timeout, :idle_timeout_ms). Functions and pids can't be saved, so pass :subscriber and :handlers to the reaper; it forwards them to every session it restarts. Without them, a revived session runs the default handlers and has no subscriber to tell about a pending approval.

Run one reaper per node (host supervision tree). Concurrent reapers on multiple survivors are safe: the registry's atomic :via registration means only one start wins; a loser's {:already_started, _} is treated as a successful start (the shared Turn.Start.start_server/6 maps it to {:ok, pid}), so nothing is logged or retried.