[codex] Stop binding permission boundaries to trust edges

[InfoSecHack]

Summary

• Stops �ind_permission_boundaries from attaching permission-boundary bindings to trust edges.
• Keeps permission-boundary bindings on permission edges where the constrained principal is the edge source.
• Updates comments/docstrings and tests that previously described boundaries as dst-side trust-edge constraints.
• Keeps cross-account trust defensive filtering and adds a malformed-graph regression proving non-SCP trust-edge bindings are ignored.

Root Cause

Permission boundaries constrain a principal's effective permissions after credentials exist. They do not constrain who can assume a role through its trust policy. Binding a target role's boundary to trust edges was semantically misleading and risky for future reasoners, even though cross_account_trust already filtered non-SCP trust-edge bindings defensively.

Validation

• python -m pytest -q tests/resolver/test_permission_boundary_binder.py tests/integration/test_full_pipeline_reasoner_verdicts.py
• python -m pytest -q tests/test_permission_boundary.py tests/resolver/test_permission_boundary_binder.py tests/integration/test_full_pipeline_reasoner_verdicts.py tests/test_cross_account_reasoner.py tests/test_passrole_lambda_reasoner.py tests/test_passrole_ecs_reasoner.py tests/test_golden_findings.py
• ./scripts/check.sh
• ./scripts/test_fast.sh
• git diff --check
• account/ARN hygiene scans
• Terraform/raw artifact scan