fix(deep): preserve checkpoint authority during recovery

[mldangelo-oai]

Summary

Stopped Deep scans can resurrect a rejected candidate by reading an older result file, or change their recovery input after a publication write fails. Read the committed checkpoint head and freeze that selection so replay preserves the accepted evidence and latest disposition.

Changes

• Prefer the committed discovery checkpoint over older current and archived results; order archived attempts numerically.
• Freeze selected checkpoint paths alongside existing retained source digests before publication. Explicit recovery can adopt a later head; ordinary replay uses the frozen selection.
• Preserve accepted partial evidence, original failure state and legacy results without checkpoint-head metadata.
• Add migration 47 for the nullable frozen-head mapping; existing migrations remain unchanged. No public CLI change.

Testing

On bd70d41478d06331f297decd171d0d11d7fd439f:

• The schema-capable parent has 10 regression failures and 10 passing controls; the same tests pass all 20 cases after the correction.
• Affected Python suites: 552 passed, two skips and 55 subtests. Four affected SDK suites: 82 passed, 584 assertions. Four affected MCP suites passed.
• Installed SDK and physically detached plugin: 20 cases passed, including five new-process replays and public contract reads.
• All five required portable checks passed: Ruff lint/format, SDK build:ci, plugin source compatibility and its tests. SDK/plugin builds passed.
• Tests exercise real output-write failure after findings reach disk, restoration of prior outputs, changed or missing heads, newer rejection, partial evidence and explicit recovery.
• Independent composition with the publication fence passes 44 targeted cases; a second verification passes 45 with one additional existing control. Current-main source composition is clean. Native platform CI remains pending.

Risk and rollout

This draft contains both compatible readers and the frozen-head writer. Establish a reader-capable rollback artifact before enabling new writes; the schema-only commit is insufficient. Coordinate migration 47 with other open migration work and keep its identity unchanged.

The change preserves original stopped state and source digests. Legacy records without frozen-head metadata retain their existing fallback. Keep the draft unmerged until reader-first ordering and platform acceptance are complete.

Public disclosure review

• No customer, partner, prospect, or user identities, data, or identifying details are included.
• No credentials, personal data, private source, scan findings, or nonpublic links or tickets are included.
• I reviewed the branch name, title, description, commits, changes, comments, logs, screenshots, attachments, and links for public disclosure.

[mldangelo-oai]

@codex review

Please review the current head, bd70d41478d06331f297decd171d0d11d7fd439f.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-12T10:01:41.313322Z	bd70d41	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Hooray!

Reviewed commit: bd70d41478

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".