Text this: LOGIC: LLM-originated guidance for internal cognitive improvement of small language models in stance detection