Fix duplicated Codex token counting in forked sessions

[sszzz830]

Summary

This fixes double counting for Codex token usage when a session is forked.

Forked Codex sessions get a new session_id, but they inherit the parent session's cumulative token_count history. The existing scanner treated the forked session as fully independent and counted that inherited prefix again, which inflated token and cost totals.

What changed

• Read session_meta.forked_from_id from Codex session logs
• Track raw token_count entries per session during scanning
• When a child session is forked from a parent, compare the child's raw prefix against the parent's existing prefix
• Skip only the inherited prefix that exactly matches the parent history
• Continue counting all new usage generated after the fork in both parent and child sessions
• Bump the Codex cost usage cache artifact version from v2 to v3 so previously overcounted cached results are invalidated

Why this approach

This is intentionally strict and narrow:

• It only applies to explicit Codex forks with forked_from_id
• It does not dedupe unrelated sessions that merely happen to have similar totals
• It preserves post-fork usage correctly for both branches
• It avoids over-deduping by only removing the exact inherited prefix

Validation

Validated against forked-session scenarios to confirm that:

• inherited parent usage is not counted twice
• new child usage after the fork is still counted
• new parent usage after the fork is still counted
• incremental rescans still work correctly
• the Codex cache now rolls to codex-v3.json

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cf87e675d4

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Invalidate fork-deduped child cache when parent disappears

This code computes child usage with parentUsageEntries and then caches the deduped days, but it does not persist any dependency on the parent session/file. On later refreshes, unchanged child files take the mtime/size fast path and are not reparsed, so if the parent file is removed from the scan set, the end-of-scan cleanup subtracts the parent totals while the child still excludes the inherited prefix, causing an undercount until the child changes or a force rescan is run.

Useful? React with 👍 / 👎.