CorrectionLM: Self-Corrections with SLM for Dialogue State Tracking