[codex] let later shared reads join a running read window

[xli-oai]

Summary

• replace fixed shared-read batches with per-key queue state in the app-server scheduler
• allow later shared reads to start while earlier shared reads are still running when no writer is queued
• preserve writer fairness once an exclusive request is waiting

Why

The existing scheduler only formed one batch of leading shared reads, then waited for that whole batch to finish before looking at the queue again. That meant a later plugin/list arriving behind an already-running skills/list still waited even though both requests were safe shared reads.

Behavior change

A shared read now joins the active read window immediately when there is no running writer and no queued writer. Exclusive requests still wait for all active reads to finish, and once a writer is queued, later reads stay behind it.

Validation

• just fmt
• just fix -p codex-app-server
• cargo test -p codex-app-server
  • scheduler unit coverage passed, including the new regression test
  • the broader crate run still hit existing environment-sensitive failures in this local setup (sandbox-exec command-exec tests, one skills watcher timeout, and one websocket-dependent turn interrupt test)
• manual dev desktop verification: later plugin/list requests completed while an earlier long skills/list was still in flight, which is the behavior this change is intended to enable

[xli-oai]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Keep them coming!

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".