feat(server): add Experience and Skill review UI

[Teingi]

Which issue or RFC does this PR close?

Closes #1302.

Implements the bilingual Experience and Skill Review Page RFC included in this PR.

Rationale for this change

The Dashboard exposed review counts, but reviewers still needed CLI or MCP calls to inspect evidence, revise Candidates, make decisions, and publish approved Skills. Agent-native Skills also lived in separate host-local directories without one scoped view of managed and external content.

This change adds a visible governance loop while keeping the authority boundaries explicit: approval creates an immutable Artifact Revision, and publication to an Agent target remains a separate action.

What changes are included in this PR?

• Add an authenticated Review Inbox at /reviews for scoped Experience and Skill inspection, revision, approval, rejection, and optimistic-concurrency recovery.
• Add a Skills Library at /skills for managed and external Skills, search and authority filters, delivery status, and successor-revision creation.
• Add a unified AgentSkillTarget abstraction with Codex and Claude Code discovery and managed publication support.
• Preserve legacy codex_roots configuration as a compatibility form.
• Publish only exact approved Skill Revisions to allowlisted targets, detect drift and conflicts, and never overwrite foreign content.
• Extend the OpenAPI contract and generated models for Claude Code external Skill registrations.
• Add bilingual RFC and reference documentation, responsive UI styles, locale coverage, and focused server, registry, projection, and browser-surface tests.

Are there any user-facing changes?

Yes. The Server UI gains Review and Skills navigation entries, and Server configuration can declare Codex or Claude Code Agent targets. Existing codex_roots configuration continues to work. There are no breaking public API or persisted-format changes.

How was this change tested?

• make check
• make test — 634 passed, 13 skipped on the final merged tree
• make api-generate
• make contract-test — 29 passed
• make docs-test
• make build
• Inspected the built wheel for the Review and Skills templates, JavaScript, and shared CSS assets.
• Ran real Firefox flows for Experience revise/approve/reject, stale-version draft recovery, Skill revise/approve, Codex and Claude Code publication and discovery, Skills Library search/filtering, and successor Candidate creation.
• Verified English and Chinese locales, dark mode, a 390 px viewport, searchable keyboard-operated scope selection, and zero browser console errors.
• Verified HTML-like untrusted Candidate content remains inert in lists, details, forms, and conflict recovery.

AI usage statement

OpenAI Codex (GPT-5) was used for design iteration, implementation, code review, and test execution.

[Teingi]

[Teingi]