Preserve distinct coding-agent conversation identities during import

[witbrock]

Problem and result

Machine-wide imports collapsed Codex child rollouts into their ancestors when inherited metadata followed the child header. Gemini A2A runs also reused a2a-server, merging unrelated conversations. Preserve the first Codex session identity and distinguish A2A runs by stable start time, while retaining original identifiers and raw bytes. Include archived Codex sessions in default discovery.

Validation

• 28 targeted parser, bulk-import and authenticated-route tests pass, including sibling separation and stable identity as logs grow.
• Read-only replay of 36,007 real Mac sources: 35,972 readable conversations, 35,972 distinct identities, no incomplete visible projections. The remaining 35 files have no supported visible messages.
• Existing DGX pilot imports verify private raw custody, canonical history readback and unchanged repeat imports. Full bulk activation waits for this repair.

Merge decision changes — ready: the observed identity-collision failure is fixed with targeted regression and real-source evidence. Missing Gemini A2A start times use content-hash identity and cannot promise append reconciliation; documented explicitly.

Native task: #V#task_preserve_distinct_codingagent_conversations_during_31afcfc2.