Domain-aligned LLM framework for trustworthy scientific Q/A via query reformulation retrieval-augmented generation