Fairness-Driven LLM-based Causal Discovery with Active Learning and Dynamic Scoring