fix: time out stalled LLM streams

[askie]

Related Issue

Related to #1798.

Supersedes #1799.

Problem

Provider request timeouts stop protecting the call after response headers arrive. If the streamed response then stops producing normalized message parts, generate() can remain blocked in iterator.next() indefinitely. Existing abort checks only run between parts, so they cannot recover a turn from this state.

What changed

• Add a per-call streamIdleTimeoutMs driver option with a three-minute default in both current generation contracts.
• Race every streamed iterator.next() against the idle deadline and the caller abort signal, covering the first part, subsequent parts, and stream completion.
• Abort the provider transport through a merged signal and throw a retryable StreamIdleTimeoutError with provider, model, elapsed-time, and trace metadata.
• Keep caller cancellation as the standard non-retryable AbortError.
• Perform stream and iterator cleanup on a best-effort basis without awaiting provider cleanup promises that may also be stalled.
• Strip the driver-only idle deadline before forwarding options to provider adapters.
• Add contract tests for first-part and inter-part stalls, retry classification, transport abort, non-blocking cleanup, caller abort precedence, and option forwarding.
• Export the new error from the @moonshot-ai/kosong root entry and add a CLI patch changeset.

No environment variable or new user-facing configuration surface is introduced, so the documentation audit found no user documentation update to make.

Checklist

• I have read the CONTRIBUTING document.
• I have linked a related issue, or explained the problem above.
• I have added tests that prove my feature works.
• Ran gen-changesets skill, or this PR needs no changeset.
• Ran gen-docs skill, or this PR needs no doc update.

[changeset-bot]

🦋 Changeset detected

Latest commit: ea96d22

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@moonshot-ai/kimi-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ea96d22866

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]

Bound the idle delay before passing it to setTimeout

When a caller supplies a finite streamIdleTimeoutMs above Node's maximum timer delay (2,147,483,647 ms), resolveStreamIdleTimeoutMs accepts it, but Node 24 emits TimeoutOverflowWarning and schedules this timer for 1 ms. A caller using a large duration to effectively relax the deadline will therefore have every stream fail almost immediately. Reject or clamp out-of-range values before scheduling; the mirrored agent-core-v2 implementation needs the same correction.

Useful? React with 👍 / 👎.