Expose a dispatch matrix blind spot with source mutation probes

[aruintelligence]

What changed

• Rerun the 192-vector matrix against seven exact, single-guard substitutions of the real reference boundary. Six substitutions change rows; the resource-match substitution changes none.
• Add a direct local resource-substitution regression probe: a policy for team@example.test must block a proposed outside@example.test callback. The mutant reaches one callback.
• Publish a deterministic source-bound mutation report and a visual finding in the dispatch atlas. Keep the finite project-authored scope explicit.
• Run the report and regression test in the action-boundary CI workflow.

Verification

• Node boundary/vector checks, Python six-vector and 192-vector oracles
• Mutation report --check and seven focused Node tests
• Local link and deployment-surface checks, clean diff

No provider or external message callback is used.