Document the agent-eval workflow and add it to the release checklist

[ecomodeller]

Part of the agent-documentation plan (#995).

Depends on #1005.

Without this the eval runs once and never again, and AGENTS.md drifts out of sync with the library at exactly the rate the API changes.

• Add a short CONTRIBUTING.md section: what AGENTS.md is for, that its snippets are tested, and how to run the eval.
• Add "run just agent-eval and commit the scorecard" to the release checklist, so drift is caught at a natural cadence.
• Consider an ADR recording the decision to ship agent-facing documentation inside the package, and why the wheel is the primary channel rather than the docs site.

Acceptance: a contributor can find out how to run the eval without asking.