Reduce parallel recovery growth increment to 0.1

[hemanth-asirvatham]

Motivation

• Reduce how aggressively the concurrency cap ramps up after sustained success to make parallel recovery more conservative and avoid overshooting during transient recovery.

Description

• Change the parallel recovery increment multiplier from 0.15 to 0.1 in src/gabriel/utils/openai_utils.py so increment = max(1, int(math.ceil(max(concurrency_cap * 0.1, 2)))); also verified that rate-limit and connection errors use the existing retry/backoff behavior via _handle_rate_limit_error and the APIConnectionError branch which enqueue retries through _maybe_retry when attempts remain.

Testing

• No automated tests were executed for this change (no test run requested).

---

Codex Task

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}