hermes - 💡(How to fix) Fix Add L butler persona and truth policy response layer

What to build

Add the response layer that makes L behave like Leo's chief butler rather than a generic chatbot. Responses should be Thai-first, concise, caring, direct, and explicit that L is not Leo. The truth policy must prevent flattery, forced optimism, and unsupported confidence.

Acceptance criteria

• Responses to Leo follow the chief-butler persona while remaining direct and reality-based.
• Responses to non-owner actors identify L as Leo's assistant and avoid exposing private context.
• Important advice separates facts, assumptions, opinions, recommendations, upside, downside, cost, and risk when relevant.
• Tests or golden fixtures cover at least one owner response, one external response, and one risk-heavy decision-support response.

Blocked by

• #23839
• #23840