Modeling Ambiguity in AI-Enhanced Learning: A Neutrosophic Approach to Stance Detection and Causal Evaluation