fix(openai): preserve JSON instructions in Codex OAuth input

[ddnio]

Summary

• map Codex OAuth input items with role: "system" to role: "developer" instead of dropping them
• keep mirroring extracted system text into top-level instructions for the ChatGPT internal Codex endpoint
• add regression coverage for Responses JSON object requests where the only json instruction is in the system message

Why

The previous transform removed system messages from input after copying their text into instructions. That can break OpenAI Responses JSON object validation because the Responses API requires the word json to appear in the input messages when text.format.type is json_object.

Keeping the message in input as developer preserves the JSON-mode instruction while still avoiding the unsupported system role for the Codex OAuth upstream.

Validation

• go test ./internal/service -run 'TestApplyCodexOAuthTransform|TestExtractSystemMessagesFromInput|TestIsInstructionsEmpty' -count=1
• go test ./internal/service -count=1
• git diff --check upstream/main...HEAD

[github-actions]

Thank you for your contribution! Before we can merge this PR, we need you to sign our Contributor License Agreement (CLA).

To sign, please reply with the following comment:

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

You only need to sign once — it will be valid for all your future contributions to this project.

---

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

---

欧阳丹 seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}