Add candidate validator, model tests, and defense-in-depth CI

[atimics]

Scope

Implement the non-authoritative validation layer for the v1 constitutional proposal:

• standard-library schema and semantic validator
• canonical human rendering drift check
• policy-model adversarial unit tests
• least-privilege, immutable-action candidate CI
• base-revision validator as defense in depth

Trust boundary

This tooling must never ratify, activate, or authorize from caller-supplied identity or signature metadata. Passing candidate CI is not independent evidence. External verification against the active predecessor remains required.

Acceptance

• all schemas are applied and unsupported keywords fail closed
• structural decision helpers return only deny or review, never allow
• L0/proposed posture grants no capability
• semantic/path risk escalation is tested
• unsigned approval and activation metadata never authorizes
• workflow permissions are exactly contents: read
• branch remains within the workspace change budget

Stack this change on issue #1's normative policy branch until that proposal merges.