feat(framework)!: remove evals system end-to-end

[blafourcade]

What

Removes the evals concept across the whole framework end-to-end.

Why

The evals fixtures were not isolated test data: every evals/scenarios.json doubled as the BM25 training corpus for the auto-routing hook. Removing evals therefore means removing the routing engine they fed, the authoring path that produced them, and the gates/docs that referenced them.

Scope

• Fixtures — delete all 31 plugins/*/skills/*/evals/scenarios.json.
• Routing engine — delete aidd-context/hooks/routing/ (bin, lib, routing-hint, tests); unwire its SessionStart + UserPromptSubmit hooks in hooks.json (update_memory.js kept).
• Authoring — drop 02-design-evals action + evals-template asset from 03-context-generate; renumber the skills action chain to 01-05 and the tool-resolution gate to R10.
• Gates/docs — remove evals-validity from lefthook.yml; scrub evals references from docs, memory, and per-plugin CATALOG.md.

Out of scope / preserved

• CHANGELOG.md files untouched (release-please owns them).
• False positives kept: evaluate/evaluation prose and the eval() injection warning in the security rule.

⚠️ Breaking change

Auto-routing is removed. Skills become manual-invoke only; the prompt-to-skill routing hint no longer runs.

Verification

• 0 scenarios.json / evals/ left on disk; dangling-ref sweep clean.
• hooks.json valid JSON; lefthook.yml valid YAML.
• 05-review ran: 2 renumber leftovers found (stale 06-validate include, R11 gate gap) and fixed.
• 77 files changed, 58 insertions, 2280 deletions.

🤖 Generated with Claude Code