test: close test framework parity gaps

[Pluze]

Goal and scope

• Goal: close the behavioral parity debt from the catalog cutover and remove migration-era bypasses without restoring legacy runners, wrappers, selectors, or test-owned product APIs.
• In scope: owner-first App and framework evidence, path-isolated App conformance, complete changed-file fallback, aggregated failure reporting, production-source routing guardrails, bounded ManualChecks, structural GUI conformance, representative App workflows, repository hygiene, and contract-facade compatibility behavior.
• Intentionally out of scope: native dialog feel, pointer feel, visual design review, real lab data, and scientific interpretation; these remain explicit non-passing manual checks.

User-visible behavior

• Before: a green catalog did not prove the high-value legacy behavior set; GUI evidence was largely startup smoke; unknown changes omitted GUI/path-isolation evidence; an isolation helper started a second MATLAB; and an agent skill still exposed the deleted runner model.
• After: catalog, plan, and run remain the only execution path; all audited App rows have current direct or bounded workflow evidence; every production MATLAB source has an explicit evidence location; path isolation runs inside the unified executor; and active guidance/automation contains no legacy runner wrapper.
• Compatibility / upgrade note: test and maintainer-tooling migration only. App algorithms, saved project formats, exports, and public LabKit APIs are unchanged.

Validation

• I ran the smallest relevant local validation from docs/development/maintain-and-release/testing.md.
• I ran the required final gate for this handoff, or this PR is not ready for final review.
• Automated validation was not run; reason and risk are recorded below.

Commands and results:

framework/contract source: 4 passed
system/build system: 30 passed
system/repository system: 3 passed
apps/conformance hidden-gui: 21 passed
isolated profile: 1 aggregate identity passed for all 21 public Apps
docs + docsCheck: 218 narrative pages, 149 API pages, 372 generated files current
changedFast full branch gate: 298 passed (258 headless, 39 hidden-GUI, 1 path-isolated)
FocusStack fixture changedFast: 2 passed (presentation + hidden-GUI workflow)
PR CI: all 9 Linux, macOS, and Windows jobs passed

Manual / native behavior not verified:

• Native dialogs, pointer feel, visual design quality, real-data suitability, and scientific interpretation. Mapped layouts emit concrete non-passing ManualChecks for these boundaries.

Documentation and boundaries

• Human docs were updated where user-facing behavior or public contracts changed.
• Agent guidance was updated where execution, ownership, or governance rules changed.
• Docs and agent guidance were left unchanged because their contracts were preserved.
• Versioned component docs and a related docs/history/records/**/*.md record were updated where required.
• App-specific workflow logic stayed in the owning app.
• Reusable +labkit changes, if any, are domain-neutral and app-facing.

Boundary decision:

• No product-facing API was added. The temporary test-only DefinitionInspector method was removed; the conformance spec uses the existing private inspection boundary.

Delivery state

• The branch is committed and pushed.
• Required GitHub checks are linked below, or are pending:

CI / review links:

• Continuous Integration for 41cca68b: https://github.com/Pluze/LabKit-MATLAB-Workbench/actions/runs/30072490315 (all 9 jobs passed).

Data hygiene

• This PR does not add raw lab files, identifying file names, subject names, device serials, local absolute paths, timestamps, or private experiment labels.

Unverified behavior

• Automated hidden-GUI checks do not prove native dialog behavior, pointer feel, visual quality, real-data suitability, or scientific interpretation.