docs(agent-docs): enforce canonical AGENTS symlinks

[caio-pizzol]

Follow-up to #3296. Standardizes the AGENTS.md / CLAUDE.md symlink direction across the repo.

Every nested location already used AGENTS.md as the real file and CLAUDE.md as the symlink. Only the repo root inverted that pattern. This PR flips root to match, with AGENTS.md as the canonical file and CLAUDE.md preserved for Claude Code compatibility.

• Policy: adds a canonical symlink direction rule under Placement.
• L1 scanner: reports symlinked pairs with the wrong canonical file as linked-inverted deterministic findings.
• Root files: flips AGENTS.md to the real 85-line file and CLAUDE.md to a symlink.

Verified locally: L1 audit no longer reports the root pair as inverted; the 12 linked AGENTS.md / CLAUDE.md pairs classify as canonical: AGENTS.md, with packages/superdoc intentionally different and allowlisted.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0047512d32

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Include pair findings in the full audit summary

When an inverted symlink pair is found during a full weekly/dispatch audit with ANTHROPIC_API_KEY available, this helper marks the pair as review-worthy only for renderL1Markdown; flaggedForReview(scan) still returns only files, and .github/scripts/agent-docs-audit.mjs prints renderMarkdown(report) rather than the L1 markdown on the non-skip path. As a result, linked-inverted pairs are uploaded in /tmp/agent-docs-audit-l1.md but are absent from the Step Summary/JSON that reviewers normally see, so the new canonical-direction finding is easy to miss unless AI is skipped.

Useful? React with 👍 / 👎.