Write five eve evals for the enquiries agent

The agent is two tools and an instructions file; the evals are where the work is. Every grader is a pure function in evals/enquiries/graders.ts, so the same code runs inside eve eval and can be called from any script with no server. The agent's model comes from .env.local: an OpenRouter model on your own key, a local Ollama model, or a fixture model that plays a careful agent from a script. The decision that matters most is running the suite on the fixture model first, with planted faults, so you know each eval passes on a correct output and fails on the mistake it targets before any real model is involved.