feat: compress rollout files

[jif-oai]

Compress rollout files using ZSTD. I chose ZSTD because you can append to the compressed payload

On my local ~/.codex this is a 76% compression

[jif-oai]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

codex/codex-rs/exec/src/lib.rs

Line 1127 in 74bd00f

let text = tokio::fs::read_to_string(path).await.ok()?;

Decode compressed rollouts before parsing latest cwd

New rollouts are now created as .jsonl.zst, but this function still reads the file with tokio::fs::read_to_string. For compressed files that returns an invalid UTF-8 error, so parse_latest_turn_context_cwd returns None and latest_thread_cwd falls back to stale thread metadata. That can make resume --last/cwd filtering pick the wrong thread after in-session cwd changes.

---

codex/codex-rs/tui/src/lib.rs

Line 1422 in 74bd00f

let text = tokio::fs::read_to_string(path).await.ok()?;

Read zstd rollout when deriving TUI TurnContext state

read_latest_turn_context now fails on newly compressed rollout files because it uses read_to_string on .jsonl.zst. When state DB is unavailable, read_session_model/read_session_cwd can no longer recover the latest TurnContext, so resume/fork flows may silently fall back to defaults or initial session metadata instead of the last persisted context.

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".

[jif-oai]

Indeed I have to fix backward compatibility with resume