feat: set PR review reasoning effort to high

[wkj2333666]

Summary

• Set automated PR review reasoning effort to high by default.
• Send the protocol-specific reasoning field for Responses and Chat Completions.
• Document the OpenAI GPT-5.6 Sol configuration.

Validation

• 74 passed, 9 subtests passed
• Focused PR-review suite: 20 passed
• Workflow YAML parsing, Python compilation, and git diff --check passed.

[wkj2333666]

自动 AI 审阅完成：decision=pass，无 P0/P1，存在 1 条 P2 建议。建议补齐这一兼容路径：初始 502/503/504 后，无 reasoning 的 strict retry 如果返回 400/422，也应继续回退到 legacy json_object；同时增加 504 -> 400/422 -> legacy success 的回归测试。该 PR 为当前账号创建的 draft，GitHub 不允许自我 approve，因此保留为非阻塞评论。运行记录：https://github.com/SAST-agent/AgentBenchFramework/actions/runs/30259881091