[Bug]: 2.2.0-beta.5 QwenPaw Hub fails to connect to the model service in some cases.

## QwenPaw Version

2.2.0-beta.5

## Description

To try out QwenPaw Hub, I updated to version 2.2.0-beta.5. After testing, I found that the cloud API service is accessible, but connecting to a specific local API seems to fail: `http://127.0.0.1:8088/v1`, this also includes LAN addresses such as `192.168.x.x` that match the local machine's address.

QwenPaw Hub: Listens on the address `0.0.0.0`, on port `8000`, and the access address is `http://192.168.x.x:8000`.

## Component(s) Affected

- [x] Core / Backend (app, agents, config, providers, utils, local_models)
- [ ] Console (frontend web UI)
- [ ] Channels (DingTalk, Feishu, QQ, Discord, iMessage, etc.)
- [ ] Skills
- [ ] CLI
- [ ] Documentation (website)
- [ ] Tests
- [ ] CI/CD
- [ ] Scripts / Deploy

## Environment

- **QwenPaw version:** 2.2.0-beta.5
- **OS:** MacOS 26.5.2
- **Install method:** one-line install

## Steps to Reproduce

When adding a custom model: `http://127.0.0.1:8088/v1`, I can successfully add the DeepSeek API and LM Studio on the same LAN in QwenPaw Hub, but I cannot access the model on the local port.

After testing, the model on the local port is functional, and it works normally in a standalone QwenPaw instance. However, in the QwenPaw instance launched by QwenPaw Hub, it is completely unusable under the same settings, displaying: `Connection error.`

## Logs / Screenshots

```
2026-09-01 02:12:18 | WARNING | /Users/admin/.qwenpaw/venv/lib/python3.12/site-packages/qwenpaw/providers/provider_manager_discovery.py:410 | Model discovery failed; using static fallback
```