[Feature]: 设置-模型服务"测试连接"支持不填 Model ID 即可验证连通性

[HuaXiawithMoon]

Problem

当前"设置 → 模型服务"中的"测试连接"按钮调用 onboarding.provider.probe RPC，底层 probe_llm_provider() 硬性要求 model id（空则直接 raise ValueError）。

这意味着用户必须先填写一个正确的 model id 才能验证 API Key / Base URL / 代理是否可达。对新用户来说这是一个鸡生蛋的问题：不知道 provider 支持哪些 model → 无法测试连接 → 无法确认配置是否正确 → 无法获取 model 列表。

Proposed behavior

当 model 字段为空时，"测试连接"改为调用已有的 discover_provider_models()（即 GET /v1/models 列表接口）来验证连通性，而非发送 1-token chat completion。

Area

Web UI

Alternatives considered

No response

[lifelmy]

Opened a draft PR implementing this: #1529.

Following the proposal, when the model field is empty the onboarding.provider.probe RPC now verifies reachability through the model-list endpoint (discover_provider_models() → GET /v1/models) instead of a one-token chat completion, and the Web UI no longer disables Test connection until a model id is entered (for the primary-provider draft). The listing outcome is mapped onto the existing probe envelope, so the wire contract is unchanged; chat-only timing fields stay null when no chat round-trip happens. Stored/draft profile probes keep resolving a concrete model.

Local checks are green: ruff, pytest (onboarding/contracts/gateway 1196 passed incl. two new regression tests), mypy, the affected vitest suites, npm run build/typecheck, and uv build --wheel.

I use a coding assistant to help implement, and I review and take responsibility for the final change.