Track accepted frozen React frontend-matrix failures

Owner: Showcase maintainers

Frozen base: dd20114b015ae6f76fe97b5dff5a295d2ca8c94e
Workflow run: https://github.com/CopilotKit/CopilotKit/actions/runs/29943191045

The complete 660-cell React baseline ran against branch-local integration images and fixtures. Ten cells passed and 650 cells failed. All 650 failed cells stopped in the deterministic probe layer: 730 sub-probes reported conversation-error, 710 before any turn completed. The aggregate recorded no browser page errors.

Classification: probe. The current D5 probe and frozen fixture contract does not establish a passing assertion path for these cells. This issue owns the accepted failures until the probe and fixture contracts are aligned.

The Angular parity gate must still block any new React regression and every PR-React-pass / PR-Angular-fail pair. It may report, without blocking, only the exact base-failing cells listed in the committed immutable aggregate and policy.

Acceptance:

• Reproduce from the immutable aggregate attached to run 29943191045.
• Split product, integration, fixture, or infrastructure defects into narrower issues if later evidence changes the class.
• Remove each policy entry when its frozen-base comparison no longer needs this exception.
• Do not add wildcard exceptions.

Start with the immutable aggregate from workflow run 29943191045, then inspect the D5 probe, frozen fixture contract, and React/Angular parity policy described in the issue. Reproduce the accepted failures, separate any changed defect classes into narrower issues, and consider the work done when only exact frozen-base cells remain excepted and no wildcard exceptions are added.