[Feature] 为什么local_chat方式不支持qwen3-moe模型呢？

### 检查清单

- [x] 1. 如果您提出的不是新功能而是问题，请在讨论区发起讨论 https://github.com/kvcache-ai/ktransformers/discussions。否则该 issue 将被关闭
- [x] 2. 为方便社区交流，我将使用中文/英文或附上英文/中文翻译（如使用其他语言）。未附带翻译的非英文/中文内容可能会被关闭

### 需求背景

有没有人能解答一下？

### 相关资源

_No response_

The issue does not name a file, test, entry point, or concrete acceptance criteria. Start by tracing how the local_chat mode handles model support and compare that path with Qwen3-MoE requirements; completion criteria would need to be agreed before implementation begins.

Written by the indexing model from the issue text.