Root supervisor for the Arrea engine layer.

Strategy: :rest_for_one

Children are ordered by dependency: each process depends on the previous ones, and only the dependent processes are restarted when one fails.

1. Registry (Arrea.Registry)              — Workers
2. Registry (Arrea.CircuitBreaker.Registry) — Circuit breakers
3. Arrea.Monitor                           — Depende de los registries
4. Arrea.Leader                            — Depende de Monitor y registries
5. Arrea.WorkerSupervisor                  — Depende de Leader y registries

Con :rest_for_one:

• Si falla un Registry → reinicia todo (raro; los registries son muy estables)
• Si falla Monitor → reinicia Monitor + Leader + WorkerSupervisor (batches activos se pierden)
• Si falla Leader → reinicia solo Leader + WorkerSupervisor (Monitor y registries intactos)
• Si falla WorkerSupervisor → reinicia solo WorkerSupervisor (impacto mínimo)

Esto es significativamente mejor que :one_for_all, donde cualquier fallo (incluyendo el del WorkerSupervisor) reiniciaba todos los procesos.