Increase notification health adapter coverage for HealthServer.Start

[Copilot]

This PR closes a meaningful coverage gap in the notification HTTP health adapter by adding missing lifecycle tests around server startup and shutdown behavior. The previous suite validated /healthz responses but did not exercise HealthServer.Start.

• Coverage gap addressed (Start lifecycle)

  • Adds a test for invalid bind address to assert Start surfaces listener startup errors.
  • Adds a test that starts a real ephemeral listener, confirms readiness, then cancels context and asserts graceful return.

• Behavioral surface now covered

  • Startup failure path (bad address).
  • Runtime shutdown path (context cancellation while serving).
  • Existing handler behavior remains covered (200 ok, 503 queue connection closed, 405 for non-GET).

• Representative test pattern

  addr := mustFreeTCPAddr(t)
  srv := NewHealthServer(addr, stubQueue{alive: true})
  
  ctx, cancel := context.WithCancel(context.Background())
  errCh := make(chan error, 1)
  go func() { errCh <- srv.Start(ctx) }()
  
  // wait until /healthz is reachable, then cancel
  cancel()
  
  if err := <-errCh; err != nil {
      t.Fatalf("Start returned error after context cancel: %v", err)
  }