From Imitation to Introspection: Probing Self-Consciousness in Language Models