feat(guardian): stabilize iterative review experiments

[Cyoyuu]

Summary

Stabilize iterative Guardian experiments and preserve local-specification understanding across solver revision cycles.

Changes

• validate only effective export-ignore and export-subst attributes when materializing exact Git revision snapshots
• add explicit Codex feature controls so Guardian uses unified execution without unavailable Code Mode hosts
• configure explorer and aggregator models independently while preserving the legacy shared-model fallback
• persist a host-owned, snapshot-aware local-specification ledger with append-only audit events
• process checkpoint requests along the controller-owned commit chain and report degraded, failed, and superseded reviews truthfully
• export Guardian memory with run artifacts and add focused unit and real-run coverage

Type of Change

• Bug fix
• New feature
• Breaking change
• Documentation update
• Refactoring
• Performance improvement
• Tests

Testing

• Tests pass locally

• Added new tests for the changes

• pytest focused Guardian/execution/sandbox suite: 81 passed, 2 deselected

• flake8 on all changed Python files

• isort --profile black --check-only on all changed Python files

• black --check: all 29 changed Python files unchanged (the local process required timeout termination after reporting completion)

• 20 real DeepSWE Guardian trials completed without launcher or Code Mode host errors

• the broader unit tier progressed through the changed client and Guardian surfaces; two existing process-runner tests are excluded because one terminates the surrounding test process group

Checklist

• My code follows the project's style guidelines
• I have performed a self-review of my code
• I have commented my code, particularly in hard-to-understand areas
• My changes generate no new warnings
• Any dependent changes have been merged and published

[fishmingyu]

The immediate CI and Docs failures are repository infrastructure failures, not failures in this PR. The newly enabled full-SHA Actions policy rejects setup-miniconda@v3 and setup-uv@v8.3.2 during Set up job before checkout. #534 pins every remaining floating action and adds a regression guard. Once #534 lands, this PR needs a fresh run against updated main; rerunning the old run would retain its original workflow SHA.