Text this: OGLS-SD: On-Policy Self-Distillation with Outcome-Guided Logit Steering for LLM Reasoning