Add per-component interaction tests and CI evidence gate

[amir-ba]

Summary

candidate Storybook (exact merge SHA)
  visual-tests: pixel comparisons, no baseline updates
  interaction-tests: 71 usage checks, light + dark
    live attributes, consumer forms, slotted/nested children
    public events, state, navigation and form outcomes
    38 component reports (31 active, 7 explicit exclusions)

Fix Notification Badge callback readiness without retries or sleeps. Wait for the public rendered label after attribute changes, then activate a consumer action. Observe real callbacks synchronously, including event ownership inside a surrounding control.

The two-axis review removed visual-duplicate Dropdown Select Escape/disabled checks, Sidebar collapse/keyboard checks, and Segmented Button static-disabled coverage. Tests now cover updated dropdown options, original-instance segment enabling, inserted accordion children, changed grid data, nested-link changes, runtime editing restrictions, and native form validity/submission. No production source or image baselines changed.

Make build-pr consistent: Actions v4, Node 24, frozen/noninteractive installs and bootstrap, manifest-aware caches, read-only token, shared SHA-keyed Storybook, one worker for both browser gates, and always-upload evidence. The interaction policy also rejects removal of either theme project. Branch protection must require interaction-tests to make it a required merge gate.

Component report and review | Local commands and evidence guide

Evidence

• Before: first Actions run failed the light-theme Notification Badge callback check. Coverage included duplicate visual assertions and lacked live child/attribute scenarios.
  After: final current source passed 142 cases (71 per theme), zero failures/skips/flaky results. Badge passed 20 focused repeats; narrowed Sidebar passed 10. The preceding reviewed full repeat passed 284 cases before the final Sidebar narrowing and reporter hardening.
• Verified 38 component reports, all source SHA256 hashes, and 428 screenshot/trace/accessibility attachments. Empty text trees are valid after dismissal or fulfilled blank navigation. Generated evidence is not committed; CI uploads interaction-results- for seven days.
• Positive evidence policy plus eight rejection checks passed, including removal of each configured theme. Structured workflow assertions and locked bootstrap/workspace generation passed.
• Earlier visual comparison passed 478 cases with the same 54 legacy skips. No baseline changes or relaxed thresholds. Updated remote CI and downloaded-artifact confirmation will be recorded below.
• Known product gaps are documented, not encoded as expected behavior: Menu disabled traversal/focus restoration; Slider Home/End; Segmented Button ARIA/initial sizing; Badge keyboard activation; Accordion parent-property inheritance for late children.
• All 38 original component agents received the requested GPT-6 Luna dispatch parameter; runtime identity was not independently verified.

Merge Danger

Door: two-way

Revertible tests, docs, and CI changes; no data migration.

Blast Radius: tests

New correctness/evidence gate and updated Actions setup. No consumer API or production behavior change. Leave unmerged for review.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.