Fix openai_responses_compatible full endpoint handling

[lidaxian121]

Summary

• preserve the configured full /responses endpoint for openai_responses_compatible
• align its runtime behavior with the provider config template, which already renders a complete responses URL
• add a regression test for prefixed gateway paths

Problem

openai_responses_compatible renders its configured URL as a full responses endpoint, but OpenAIResponsesProvider::new() only preserved full /responses paths for CODEX and OPENCODE_ZEN.

That meant a sanitized configuration like:
OPENAI_URL=https://gateway.example/custom-prefix/v1
would render to:
https://gateway.example/custom-prefix/v1/responses
but runtime chat requests were rewritten to:
https://gateway.example/v1/responses

This caused 404 Not Found failures for gateways mounted under a prefixed base path, even when model discovery still worked.

Closes #2711.

Testing

• cargo test -p forge_repo openai_responses

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ amitksingh1490
❌ liyang

---

liyang 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 have signed the CLA already but the status is still pending? Let us recheck it.}