MultiModal Code-Switching: Interleaving Visual Objects into Language for Explicit Object-Level Alignment