refactor: event-driven delivery engine with channel notifications and timer-based retries

Replace the polling-based delivery engine with a fully event-driven
architecture using Go channels and goroutines:

- Webhook handler notifies engine via buffered channel after creating
  delivery records, with inline event data for payloads < 16KB
- Large payloads (>= 16KB) use pointer semantics (Body *string = nil)
  and are fetched from DB on demand, keeping channel memory bounded
- Failed retry-target deliveries schedule Go timers with exponential
  backoff; timers fire into a separate retry channel when ready
- On startup, engine scans DB once to recover interrupted deliveries
  (pending processed immediately, retrying get timers for remaining
  backoff)
- DB stores delivery status for crash recovery only, not for
  inter-component communication during normal operation
- delivery.Notifier interface decouples handlers from engine; fx wires
  *Engine as Notifier

No more periodic polling. No more wasted cycles when idle.

cmd/webhooker/main.go

@@ -39,6 +39,9 @@ func main() {
 			handlers.New,
 			middleware.New,
 			delivery.New,
+			// Wire *delivery.Engine as delivery.Notifier so the
+			// webhook handler can notify the engine of new deliveries.
+			func(e *delivery.Engine) delivery.Notifier { return e },
 			server.New,
 		),
 		fx.Invoke(func(*server.Server, *delivery.Engine) {}),