serialize queueMRFHeal(), add timeouts and avoid normal build-ups (#17886)

we expect a certain level of IOPs and latency so this is okay.

fixes other miscellaneous bugs

- such as hanging on mrfCh <- when the context is canceled
- queuing MRF heal when the context is canceled
- remove unused saveStateCh channel

cmd/notification.go

@@ -1520,15 +1520,14 @@ func (sys *NotificationSys) GetReplicationMRF(ctx context.Context, bucket, node
 			select {
 			case <-ctx.Done():
 				return err
-			default:
-				mrfCh <- e
+			case mrfCh <- e:
 			}
 		}
 		return nil
 	}(mrfCh)
 	go func(wg *sync.WaitGroup) {
 		wg.Wait()
-		defer close(mrfCh)
+		close(mrfCh)
 	}(&wg)
 	return mrfCh, nil
 }