How to keep Spring AI 2.0 agents honest — RAG advisors, typed tools, structured output with schema validation, refuse-on-empty, and eval gates before you trust the loop.