[codex] add AI participation foundation evaluator

[s133per]

What Changed

• Added a new foundation-level ai-native-ai-participation-evaluator weighted at 40% of the foundation score.
• Added six child evaluators for agent thread participation, source-control AI participation, skill activation depth, AI self-assessment loops, human follow-through, and human-AI collaboration traces.
• Updated issue and PR templates so work must record AI trigger quality, skill usage, human follow-through, skipped workflow steps, and agent pushback.
• Regenerated the foundation self-evaluation report and docs baseline to include the new AI participation dimension.
• Expanded packaging tests to enforce the new evaluator tree, weight ratio, template fields, and recent-change rubric coverage.

Intentionally Not Done

• Did not run live agent evals. This change updates evaluator contracts, templates, and persisted self-evaluation artifacts rather than validating a real agent runtime behavior change.
• Did not alter product-domain code or add hosted/runtime service behavior.

Command Evidence

• node --test tests/*.test.mjs
• pnpm self-eval:validate
• pnpm --dir .agents/skills/ai-native-eval/scripts/eval test:tool
• pnpm self-eval:render
• pnpm self-eval:check
• pnpm test
• pnpm skill-eval:contract
• git diff --check

Skill Contracts Or Rubrics Changed

• Added a new foundation group evaluator and six leaf evaluator contracts.
• Updated the foundation evaluator manifest so AI participation contributes 40% of the foundation score.
• Each new leaf evaluator scores configuration evidence and recent issue/PR/thread execution evidence, with unavailable GitHub/source-control access treated as absent evidence.

Self-Evaluation Artifact Changed

• Updated self-evaluations/foundation-20260614/report.md.
• Added six new evaluator output JSON files under self-evaluations/foundation-20260614/run/evaluators/.
• Updated README/docs baseline score to 2.8 / 10, level 2, reflecting low current execution evidence for the new AI participation rules.

Human-Only Review Decisions

• Reviewers should confirm that 40% is the intended foundation weight and that the issue/PR template prompts are strict enough for future process evidence.

Skipped Gates

• Live agent eval was intentionally deferred because no real-agent behavior path changed in this PR.