feat(skills): add per-skill evals and improve description trigger phrases

[DataBoyTX]

Summary

• 8 skills now have evals/evals.json - all user-facing skills were missing per-skill evals: graphistry, pygraphistry, pygraphistry-ai, pygraphistry-connectors, pygraphistry-core, pygraphistry-gfql, pygraphistry-visualization, graphistry-rest-api. Each has 3 positive test cases + 1 negative boundary case with typed assertions (contains/negative).
• Description frontmatter improved for all 8 skills - replaced passive "Use for..." phrasing with explicit quoted trigger phrases ("Use when asked to...", "Also triggers on..."), and added proactive-suggest clauses. Goal is to reduce undertriggering, per skill-creator best practices.
• Audit report added at docs/skill-evals-audit-2026-04.md with full findings, priority matrix, and remaining work.

What this does NOT include (deferred to follow-up)

• P4 - gfql tier sizing: pygraphistry-gfql is 232 lines and could benefit from an examples/ dir to shed the quick-reference tables. Low-risk but needs a separate review.
• P5 - description optimization loop: Running scripts/run_loop.py against trigger evals for the two router skills (graphistry, pygraphistry) to auto-optimize descriptions. Requires claude CLI and longer iteration cycle.

Test plan

• python3 scripts/ci/validate_skills.py passes (18 skills validated - confirmed locally)
• Spot-check one eval JSON per skill is well-formed
• Review audit report at docs/skill-evals-audit-2026-04.md for completeness
• Optional: run one skill through the skill-creator eval loop to verify assertion format is recognized