Add capability-native policy evidence

[pengfei-threemoonslab]

Summary

• Add a capability-backed policy subject builder and route built-in policy checks plus policy packs through it.
• Expose report v0.24 capability refs/evidence while preserving fingerprints, release gating, GitHub outputs, capability lock export, and lock diff compatibility.
• Document and schema-generate capability-native policy-pack match syntax, then refresh samples and agent instruction bundles.
• Follow-up review fixes: restore canonical read-only policy parity, keep capability_refs out of finding identity, scope packet-only field stripping to release items, and update the sensitive-field inventory.

Compatibility

• release_decision.decision remains the only release gate.
• Legacy policy-pack match.* syntax is preserved; legacy risk-tag predicates use raw legacy tags for parity.
• Anthropic policy artifacts are now recognized by shared effective-control matching for missing_*_policy selectors, matching built-in policy-check behavior. This can correct policy-pack evidence/fingerprints for Anthropic-approved tools that were previously treated as missing controls.
• Capability lock export remains schema 0.1; lock diff remains schema 0.2.
• Packet schema remains 0.6; packet JSON strips report-only capability refs only from packet release-decision item lists.

Validation

• python -m pytest -q
• python scripts/generate_schemas.py --check
• ruff check .
• PYTHONPATH=src python scripts/run_governance_benchmark.py --catalog benchmark/agent-pr-governance/cases.yaml --json
• PYTHONPATH=src AGENTS_SHIPGATE_AGENT_MODE=1 python -m agents_shipgate verify --workspace . --config shipgate.yaml --ci-mode advisory --format json (merge_verdict=mergeable)