Add RBAC migration replay fixtures

[DENGXUELIN]

/claim #1351

What changed

Adds fixture-backed migration simulation and regression evidence gates to rbac-design.

• Adds RBAC-MIG-EVID-01 through RBAC-MIG-EVID-08 for before/after access diffs, historical request replay, SoD/constraint regression, special identity coverage, ABAC attribute failure tests, pilot/cutover controls, owner sign-off, and policy evaluation logs.
• Adds migration-specific severity guidance and a Migration Simulation Evidence output table.
• Adds common pitfalls for approving target-state diagrams without replay and testing ABAC only on happy-path attributes.
• Adds benign/vulnerable JSON fixtures for an approved migration replay versus unapproved privilege additions, SoD regression, missing special identities, and ABAC fail-open behavior.

Why this PR

Existing PR #1352 is a useful Markdown edge-case implementation. This PR is intentionally structured-fixture-backed so future checks can distinguish a safe authorization migration from a clean-looking target model that grants new privilege or breaks workflow during cutover.

Validation

• git diff --check origin/main...HEAD
• git merge-tree --write-tree origin/main HEAD
• JSON parse check for both added fixtures
• Markdown fence balance for rbac-design/SKILL.md
• Marker checks for version: "1.0.1", Migration Simulation and Regression Evidence, RBAC-MIG-EVID-01 through RBAC-MIG-EVID-08, Migration Simulation Evidence, Migration Decision, Approving target-state diagrams without replay, and Testing ABAC only on happy-path attributes
• Fixture marker checks for expected_skill_decision, migration_evidence, access_diff, historical_replay, constraint_regression, special_identities, abac_attribute_failure_tests, and policy_evaluation_logs
• Added-line ASCII scan
• Added-line sensitive/public-contact pattern scan
• Remote compare verification before PR creation

Bounty tier

Requesting Improver Moderate ($100) if accepted. This adds structured local fixtures in addition to the migration simulation and regression evidence guidance requested by the issue.