Retain bounded diagnostics when OCR compatibility qualification fails

[xeonvs]

Goal

Make the OCR compatibility workflow preserve a canonical issue and bounded diagnostic artifact after a qualification failure while keeping the candidate job failed and aggregate promotion blocked.

Parent: #135. Draft implementation: #134.

Acceptance

• A failed qualification writes only schema-versioned closed phase/reason data for public use.
• Issue upsert and artifact upload run after script-level failure.
• The qualification job still concludes failure and aggregate does not run.
• A later successful rerun updates the same version marker rather than creating a duplicate.
• Raw exception text, stderr, paths, commands, provider data, and credentials do not enter the issue/status projection.

Result

The workflow now retains ocr-toolkit.compatibility-status/v1 after a failed qualification, always attempts canonical issue/artifact publication, and then explicitly restores the red job outcome. The aggregate job still requires successful qualification. Closed phase/reason compatibility is validated fail-closed, including rejection of impossible combinations; open-ended exception details remain confined to the private job log.

Implementation commits: e969e97 and rebased final-head correction 1f335ba. Release delivery remains tracked in #135 and Draft PR #134.

[github-actions]

Stable v0.8.2 delivery is verified by immutable release asset release-receipt.json in v0.8.2 (SHA-256 0ac489fa17ce376895ae591b8531358f41682df76a86c36f2bf2dbb2e71ca2b3).