Representation-Based Exploration for Language Models: From Test-Time to Post-Training