feat(codex): project bounded paginated child histories without parent context

[tudstudent]

Self-contained paginated Codex child logs were rejected even when their complete ordinal prefix and child projection were present. This validates their lineage boundary and projects the child conversation without importing inherited parent history as child activity. Catalog classification now agrees with the reader, so UUID/catalog-based conversion can use the same supported sources.

• Requires contiguous integer ordinals from zero, a complete inherited prefix and consistent child metadata/boundary. External history_base, truncated streams and invalid boundaries remain fail-closed.
• Excludes inherited parent metadata/messages/tools/compaction with explicit loss accounting; child tool linkage remains intact and results referencing only inherited calls are not synthesized as child activity.
• Retains readable correspondence addressed exactly to the child's agent path as ordinary user text. Encrypted parts and envelope/routing metadata are explicit omissions; foreign/ambiguous messages remain opaque. Canonical duplicate checks require authoritative completed-item records plus exact ID, role and text.
• Imported children are independent conversations: no decryption, recovery of an entirely encrypted task, or restoration of live parent/child orchestration is claimed.
• No dependency on OpenCode v2 or Windows follow-ups. The added native acceptance test uses the existing exact OpenCode 1.17.20 public transfer path.

Validation on this exact branch: full locked Linux suite 1,548 passed, 56 skipped; focused Codex/catalog suite 87 passed, 1 skipped; selected native 1.17.20 gate 6 passed, no skips, including child import/export with parent-private/ciphertext absence assertions. Ruff/formatting, 12 automation unit tests, wheel/sdist and both isolated installed-wheel entrypoints passed.

No native Windows or complete historical native-client matrix was run. The additional development-guide Docker recipe's basic-claude-uv:latest image is absent; the selected credential-free native gate did run. Validation used synthetic fixtures and isolated stores, no real user imports, account credentials, production changes or paid inference. Scope and exclusions are documented in docs/format-compatibility.md and docs/troubleshooting.md.

This is one of three independent, main-based follow-ups split from closed #10 following the maintainer review. @xhluca, this isolates the relevant contract and its validation for review; the other two follow-ups cover separate contracts.

The eligibility bot closed #10 because the author had 64 qualifying contributions against the required 100 older than one calendar month. Splitting does not change that eligibility or claim an exemption; this submission may receive the same automation outcome.

Sibling follow-ups: Windows cleanup: #14; OpenCode v2: #15. Each targets main independently.

[github-actions]

🔴 Contributor eligibility requirements not met for @tudstudent.

• 🟢 Account created: 2011-03-09; age: 5693 days (required: at least six calendar months).
• 🔴 Contributions dated on or before 2026-09-09: 64 (required: at least 100).

Pull request authors must have a GitHub account at least six calendar months old and at least 100 GitHub contributions dated on or before the date one calendar month before the check.

See the contributor eligibility policy.

Closing this PR because these requirements were not met. This policy is intended to reduce spam from AI agents; it is not an assessment of the quality of your change.