fix(llm): surface truncated compatible responses

[koriyoshi2041]

Summary

• map finish_reason="length" from OpenAI-compatible .create() responses to the canonical OutputTooLongError
• avoid retrying the same truncated JSON response as a generic parse failure
• add a regression proving the provider does not retry the request

This restores the existing retain auto-split path for the truncation case described in #3811.

Tests

• uv run pytest tests/test_openai_compatible_response_hardening.py -q (9 passed)
• ./scripts/hooks/lint.sh
• pre-commit hooks, including unused-code checks and lint
• git diff --check

[strix-security]

Strix Security Review

Warning

This pull request has 4 commits after the last Strix review (81257e3). Strix has not reviewed these changes.
Automatic review on push is off for this repository. To review the latest changes, tag @strix-security in a comment, or turn on re-review on push.

No security issues found.

_{Updated for 81257e3.}

---

Reviewed by Strix
_{Re-run review · Configure security review settings}