fix: handle stream interruption for OpenAI-compatible providers

[jwcrystal]

Issue for this PR

Closes #20466
Related #15393, #17680, #17574, #17307

Type of change

• Bug fix

What does this PR do?

OpenAI-compatible providers (Ollama, cloud models) can interrupt SSE streams mid-response. Previously, opencode silently accepted the truncated output as a complete response — no retry, no error, just a cut-off answer.

Three fixes:

1. retry.ts — Added SSE timeout, connection reset, abort, and stream truncation patterns to retryable() matching so interrupted streams are automatically retried.

2. message-v2.ts — Classified SSE read timed out errors as APIError(isRetryable: true) instead of Unknown, so the retry mechanism recognizes and retries chunk timeouts.

3. openai-compatible-chat-language-model.ts — When flush() is called without a finish_reason from the provider while output is still active (text/reasoning/tool-call), emit an error event. This triggers the error handling path which retries, rather than silently accepting a truncated response.

How did you verify your code works?

• 5 new retryable() test cases: SSE timeout, connection reset, aborted, stream ended unexpectedly, existing rate-limit still passes
• 1 new fromError() test case: SSE read timed out → APIError(isRetryable: true, code: SSE_TIMEOUT)
• All 51 tests across retry.test.ts and message-v2.test.ts pass

Screenshots / recordings

N/A — backend fix, no UI changes.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Based on my search, I found one potentially related PR:

PR #19116 - fix(opencode): reconnect on network disruptions (VPN switch, SSE timeout, connection reset)

This PR addresses similar network and SSE timeout issues, though it appears to focus on reconnection rather than the specific retry mechanism for incomplete responses. However, it may have overlapping concerns around SSE timeout handling and connection reset patterns.

All other results were either the current PR (21727) itself or unrelated retry/error handling fixes.

No other significant duplicate PRs found.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍