feat(models): add OpenAI Codex reasoning provider

[Luhaozhu]

Summary / 概述

• Add a dedicated OpenAI / Codex provider preset for OpenAI Responses-backed Chat Completions gateways.
• Send reasoning_effort at the request root for that provider while retaining the nested chat_template_kwargs compatibility controls.
• Mark dedicated OpenAI streams as structured-reasoning streams so ordinary response text stays in the answer when the model emits no reasoning_content.
• Preserve the existing generic OpenAI 兼容 inline <think> behavior and add regression coverage for both protocols.

Validation:

• PYTHONPATH=src/backend pytest -q src/backend/tests/llm/test_openai_provider.py src/backend/tests/orchestration/test_structured_reasoning_protocol.py src/backend/tests/api/test_model_connection_probe.py src/backend/tests/services/test_user_model_selection.py — 11 passed.
• Frontend i18n check, TypeScript production build, and targeted ESLint for src/hooks/chatStream.ts passed.
• A live streaming gateway check with reasoning_effort=max produced both ThinkingBlock and TextBlock; no reasoning text or credentials are included here.
• git diff --check passed. The repository has no root npm run preflight script, so the relevant backend and frontend checks above were used.

Related issue / 关联 Issue

Follow-up to #39.

Type of change / 变更类型

• 📖 Documentation (document/**, README*.md)
• 🧩 Example / sample (examples/**)
• 🔧 Repo meta (CI, templates, .github/**)
• Other (please describe / 请说明): Maintainer sync of an upstream provider adaptation and streaming display fix.

Checklist / 检查项

• My change only touches non-generated paths (docs / examples / repo meta). 仅改动非生成路径。
• Links and code snippets were verified. 链接与代码片段已验证。
• For bilingual docs, I updated both document/en/** and document/zh-CN/** where applicable. 双语文档已同步更新（如适用）。 Not applicable; no documentation changed.
• I read CONTRIBUTING.md. 我已阅读贡献指南。