feat: add evaluation tables and data to test skills against updates

Description

Add evaluation tables and reference data so Wingspan skills can be tested against Claude Code updates. Today there is no structured way to verify a skill still behaves correctly after a Claude Code change. Define eval cases (inputs and expected outcomes) per skill so regressions are caught automatically rather than discovered in use.

Requirements

• Define an eval data format (input scenario, expected behavior/output, pass criteria) reusable across skills.
• Store eval tables/data co-located with each skill or under a shared evals/ directory.
• Seed eval cases for the core workflow skills (/brainstorm, /plan, /build, /review).
• Provide a runner that executes the evals and reports pass/fail.
• Document how to add a new eval case when adding or changing a skill.

Additional Context

Consumed by the weekly changelog-check bot: when a skill is flagged as potentially impacted, its eval table is the gate for confirming it still works. See #199.