Wake-up delivery for low-latency dispatch.

Notifiers are accelerators, never load-bearing: polling remains the correctness floor, so a lost or unavailable notifier costs latency, not jobs. Two ship built-in:

• Belay.Notifier.Local (always on) — in-process registry pokes plus :pg broadcast when BEAM nodes are clustered with distributed Erlang. Same-node dispatch is effectively instant; clustered dispatch is a message send away.

• Belay.Notifier.Postgres (opt-in) — pg_notify wake-ups through the database, for fleets that share Postgres but not an Erlang cluster:

  {Belay,
   name: MyApp.Belay,
   storage: [adapter: :postgres, url: url],
   notifiers: [:local, :postgres],
   ...}

  The listening side uses one dedicated connection per node (Postgrex.Notifications, auto-reconnecting) — point listen_url: at a direct connection if your main URL goes through a transaction pooler. NOTIFY payloads here are tiny (a queue name or job id), and if the channel is down you fall back to busy_poll/poll_interval pickup.