[OCR compatibility] Qualify v1.9.3

[github-actions]

OCR v1.9.3 compatibility evidence

• machine result: compatible
• classification: human-review-required
• all upstream asset digests and sha256sum.txt: verified
• Linux amd64 version/help/preview/result-consumer probes: passed

Upstream release changes

• release: https://github.com/alibaba/open-code-review/releases/tag/v1.9.3
• compare: alibaba/open-code-review@v1.9.2...v1.9.3
• current tested baseline: v1.9.2
• published: 2026-08-14T03:50:37Z

## 🚀 Features

- feat(resume): add trusted resume validation and transition lineage (#786) (#845)
- feat(llm): task-scoped session affinity for prompt caching (#332)
- feat(action): render category/severity badge as a shields.io image (#882) (#885)
- feat(llmloop): add grace round after tool-request budget exhausted (#872)
- feat(cmd): add SARIF output format (#820)
- feat(cmd): add no-review cmd (#835)

## 🐛 Bug Fixes

- fix(agent): stabilize review item fingerprints (#732)
- fix(cmd): report expected arguments when positional count is wrong (#892)
- fix: make FileFilter patterns case-insensitive (#859)
- fix(LE): normalize line endings via `.gitattributes` (#858)

## Other Changes

- ci: bump Go image to 1.26.6 to clear stdlib vulnerabilities (#896)
- chore: remove Chinese doc references from retry test comments (#886)
- chore(ci): fail CI when unapproved non-English text appears in source files (#876)
- Feat/llm retry report:SDK retry-attempt observability for review (#785) (#790)
- chore: remove leftover Chinese from the Go core, CI examples and pages comments (#861)

**Full Changelog**: https://github.com/alibaba/open-code-review/compare/v1.9.2...v1.9.3

Classification reasons

• release notes contain a material or ambiguous compatibility signal

Human checklist

• Review upstream release notes and relevant commits.
• Confirm compatible or incompatible.
• Confirm tested support and recommended-version promotion.
• Route any compatibility change through protected PR and stable release gates.

Workflow evidence

• run: https://github.com/xeonvs/open-code-review-toolkit/actions/runs/31778152040
• classification: human-review-required

[xeonvs]

OCR 1.9.3 qualification is complete on merged toolkit support.

Evidence:

• Feature PR Deliver M4 policy and project guidance #83 was reviewed at head 885c25c62e62882efeed79c3e78ade485a24ad33 and squash-merged as GitHub-verified commit f7d76e704fb664b78c5ad2b4678af20d54ad5c7c; the merged tree exactly matches the reviewed tree.
• Compatibility workflow 31778152040 supplied checksum-verified qualification evidence for 1.9.3, and the committed manifest recommends 1.9.3.
• The final authorized local OCR 1.9.3 review completed all 31 selected files with no failed, waived, or reused item; all nine findings and the reviewed sibling classes have negative regression tests.
• Development workflow 31793539935 completed build, trusted publication, and hosted verification for TestPyPI 0.6.0.dev50.
• Independent PEP 691 readback found the exact artifact set:
  • wheel f09a64f445ebc7467257cf55ea135d0bce5962c82d706c33690dfc2d8746d573
  • sdist e3e928d9ae8a00907f211f9e09607b05b09e3ea8fe6a22b47da200bdf0907af2
• TestPyPI Integrity readback identifies testpypi.yml, repository xeonvs/open-code-review-toolkit, and environment testpypi-public-disclosure for both exact artifact subjects.
• Registry wheel installs and the isolated ocr-ci entry point pass on Python 3.12, 3.13, and 3.14; the registry sdist independently rebuilds, installs, and runs on Python 3.14.

This closes only OCR 1.9.3 qualification. Stable toolkit 0.6.0 delivery remains tracked separately in #81.