chore(release): bump changelog to 0.5.1

[lmeyerov]

Cut CHANGELOG.md [Development] into a v0.5.1 section. No functional changes.

Patch bump per RELEASE.md: a harness bug fix, skill-text retrievability improvements, an eval-check correction, and two benchmark packs — no breaking changes and no new skills.

Contents:

• Fixed: codex exec stdin hang that burned whole per-cell timeouts on empty responses (< /dev/null in both harness scripts).
• Added: pygraphistry_skill_evals_v1, 32 per-skill cases ported from the April audit format that nothing executed.
• Changed: skill trigger phrases (Thomas Cook, feat(skills): add per-skill evals and improve description trigger phrases #23, rebased); three buried GFQL facts made scannable after the codex run showed they weren't being retrieved.
• Removed: the dead per-skill evals/evals.json format.
• Tests: per-skill pack on sonnet (32/32 on vs 16/32 off) and the cross-harness codex follow-up on the Polars pack (6/12 vs 5/12), each with its non-comparability and overfitting caveats stated inline.

Also consolidated duplicate [Development] sections that had inherited contradictory entries from #23 — it claimed both "added evals/evals.json to 8 skills" and "removed the per-skill eval format".