Track turn timing in ChatGPT telemetry

[aibrahim-oai]

Why

This is the turn-specific slice of the ChatGPT telemetry stack. We want per-turn timing that breaks total turn time into request-start delay, sampling time, and blocking tool critical path, while keeping approval wait attributable to the turn without mixing in app-server or thread startup costs.

What changed

• extend TurnTimingState in codex-rs/core to accumulate request-start delay, sampling duration, and blocking-tool critical-path duration across a turn
• emit a TurnTimingBreakdownFact at turn completion and attach it to codex_turn_event
• roll review durations into approval_wait_duration_ms and derive finalize_duration_ms as the remainder after request start, sampling, and blocking tool time
• add reducer and timing-state coverage for the new turn timing fields and approval roll-up

Verification

• cargo check -p codex-analytics -p codex-app-server -p codex-core
• just fix -p codex-analytics -p codex-core

[aibrahim-oai]

This change is part of the following stack:

• Track app-server start in ChatGPT telemetry #24142
  • Track thread start in ChatGPT telemetry #24143
    • Track turn timing in ChatGPT telemetry #24144 ◀

_{Change managed by git-spice.}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7c8290a3ff

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".

[chatgpt-codex-connector]

Track timing for interrupted turns too

This only enqueues TurnTimingBreakdownFact on the normal completion path, but interrupted turns go through handle_task_abort, which emits TurnAborted/TurnCompleted with status Interrupted without ever calling track_turn_timing. In the common case where a user interrupts a turn after a model request has started, codex_turn_event will still be emitted by the app-server path but all the new timing fields remain absent, so the telemetry slice is systematically missing interrupted-turn timing.

Useful? React with 👍 / 👎.