fix(stt): persist interrupted batch transcripts

[ComputelessComputer]

Persist streamed batch transcript snapshots and flush them so interrupted batch jobs can recover transcript progress.

---

This is part 2 of 2 in a stack made with GitButler:

•  2  fix(stt): persist interrupted batch transcripts #5246 👈
•  1  fix(stt): pass speaker count hints #5241

---

Note

Medium Risk
Changes how progressive batch transcripts are persisted (replace-mode snapshots + explicit store flush), which could affect transcript integrity/duplication and write frequency. Also alters timeout behavior by disabling idle timeouts for non-progressive/cloud batch runs, which may change failure modes for those providers.

Overview
Progressive batch runs now persist recoverable transcript snapshots. Streamed BatchResponseStreamed segment events are flattened across channels and persisted as full replace-mode snapshots, and final handleBatchResponse also writes via replace.

Batch persistence is made safer/more durable. BatchPersistCallback gains an optional { mode: "append" | "replace" }, the default batch persist in useRunBatch honors replace (avoiding re-reading existing words/hints), wraps updates in a single transaction, and triggers a save() flush after streamed writes.

Idle-timeout monitoring is gated to progressive batches. A new core helper expects_progressive_batch (exported from listener2-core) is used by the plugin to only start the 60s idle timeout monitor when a provider/model should emit progressive batch events; tests cover the new behavior.

^{Reviewed by Cursor Bugbot for commit bf94153. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 393fc08. Configure here.}