fix(codex): publish live quota while local history scans continue

[sahiinthehood]

Latest tested candidate: 6d26110 on fix/codex-quota-refresh. GitHub's closed PR head still points to the initial candidate; reopen only after issue approval/assignment to refresh the PR to the current branch tip. The description below covers that latest candidate.

Fixes #1223 (awaiting maintainer approval and assignment before reopening).

TL;DR

Keep Codex quota updates responsive when local token-history scans take longer than the provider refresh deadline. Preserve local history processing instead of letting it prevent fresh Session/Weekly values from reaching the menu bar.

What was happening

• Codex fetched live quota successfully, then awaited pricing and all three local history scans before returning its snapshot.
• A large local archive could exceed the 120-second provider deadline repeatedly. The store kept the old snapshot, leaving the menu-bar percentage frozen even though live quota was available.

What this changes

• Give local history a five-second wait budget after fetching quota.
• Keep one in-flight scan per provider instance. A slow scan continues in the background, and a later refresh collects its completed result without starting overlapping scans.
• Publish fresh quota with an existing-style provider warning while history is pending. The store's existing last-good-history preservation remains in use.
• Use byte-range newline searches in the streaming reader and Codex parser, preserving existing record boundaries and oversized-record behavior. A release-mode synthetic 32 MiB line-boundary benchmark improved from ~0.09s to ~0.027s (not an end-to-end scan-speed claim).
• Run the scan at utility priority and limit native Codex parsing to two concurrent files.
• Keep native Codex, pi, and OpenCode history aggregation and normal authentication/network error handling.
• Document the first-launch/history delay and add regression coverage for pending scans, result collection, cancellation, fast scans, and live quota publication.

Heads-up

• A completed background scan appears on a subsequent automatic or manual refresh; this patch does not add a new UI callback or change refresh scheduling.
• A permanently stuck scanner still cannot provide new history, but it no longer blocks quota refreshes or causes repeated scan launches. Local history is machine-local, as before.
• This branch is based directly on upstream main (753e2fe), with no downstream product or branding changes. No visual layout/rendering changes.

Tests

• Full swift test --disable-sandbox -j 2: 1,318 XCTest cases, 3 explicit live-test skips, zero failures; 3 Swift Testing checks passed.
• Final wait-budget adjustment: 7 focused provider/history cases passed.
• Live/parity test flags disabled; automated authentication uses fixtures.
• Cold local validation: live quota continued refreshing while the single background scan completed and wrote its persistent cache. A subsequent automatic refresh published Today, Yesterday, 30-day history, and model breakdowns.
• Final optimized build, warm-cache relaunch: quota and local history published together in 4.65 seconds, with no history-updating warning. Process returned to ~0.1% CPU.
• The five-second history budget accommodates the measured ~2.5-second warm cache load while remaining well below the provider deadline.
• Release build and local ad-hoc signing passed. No release version changes or distributed artifacts. Local iCloud provisioning was unavailable; this was a local usage/history test.

[github-actions]

Thanks for your interest in contributing to OpenUsage!

External pull requests must reference an open issue that:

1. Has been approved by a maintainer with the approved label.
2. Is assigned to the pull request author (@sahiinthehood).

Please discuss the change on an issue first, wait for a maintainer to approve and assign it to you, then reopen this pull request with Fixes #123 in its description.

Read the contribution guidelines.