DeepSeek V4 Pro 模型导致 Manager 崩溃且无法恢复 || DeepSeek V4 Pro model causes Manager to crash and cannot be recovered

配置 deepseek-v4-pro 为 LLM 模型后，Manager 因该模型的 `reasoning_content` 字段回传逻辑崩溃。崩溃后既无法通过 Manager 切换模型（Manager 已不可用），AI 网关也无切换模型的入口，最终只能重装系统。

### 复现步骤

1. 系统 LLM 配置选择 `deepseek-v4-pro`
2. Manager 启动后处理模型的 thinking/reasoning 输出
3. Manager 崩溃
4. 尝试通过 Manager 切换模型 → 不可用
5. 尝试通过 Higress AI 网关切换模型 → 无相关配置入口
6. 只能重装

### 期望行为

1. 兼容 `reasoning_content` 字段，或自动过滤后再传递给 Manager
2. 提供「逃生通道」—— Manager 崩溃后可通过 CLI 或网关后台强制切换模型，避免重装

---
After configuring deepseek-v4-pro as an LLM model, the Manager crashed due to the model's `reasoning_content` field postback logic. After the crash, it was impossible to switch models through the Manager (Manager was no longer available), and the AI ​​gateway had no entrance to switch models. In the end, the system had to be reinstalled.

### Reproduction steps

1. Select `deepseek-v4-pro` for system LLM configuration
2. After the Manager starts, it processes the thinking/reasoning output of the model.
3. Manager crashes
4. Try to switch models through Manager → Not available
5. Try to switch models through Higress AI gateway → No relevant configuration entry
6. Can only reinstall

### Expected behavior

1. Compatible with the `reasoning_content` field, or automatically filtered and then passed to the Manager
2. Provide an "escape channel" - after the Manager crashes, the model can be forcibly switched through the CLI or gateway background to avoid reinstallation