Auto-detect models in /local add

[abdulsaheel]

/local add now hits {url}/models when you land on Model ID and opens a searchable picker of what the server serves. Works with anything OpenAI-compatible (CLIProxyAPI, Ollama, LM Studio, vLLM, llama.cpp).

If detection fails it says why and falls back to typing:

• HTTP 401 Unauthorized / HTTP 404 Not Found (e.g. URL missing /v1)
• could not connect, timed out (10s), invalid URL

Other changes:

• Fields reordered: URL, API, API Key, Model ID.
• Enter with an empty model jumps to Model ID instead of erroring the form closed.
• Esc out of the picker leaves a N found, Enter to pick hint; Enter reopens it. Custom ids still typeable.
• Fixing url/key and tabbing back re-detects; stale in-flight results are dropped.
• Edit mode doesn't auto-detect (model id is the entry key).
• API Key placeholder says optional.

Tested against a live CLIProxyAPI: picked a model, saved, chatted through it. Unit tests cover parsing, auth header, HTTP errors, unreachable and invalid URLs.

Unrelated, already on main: Shift+Tab doesn't move focus in this form, the global keymap eats it as mode toggle.

[Revantark]

Thanks for the PR @abdulsaheel
Looks good to me, But I have one issue, once resolved, it should be good to go.

1. Once I select a model in the overlay, I can't select another model again. User could have selected a wrong model previously.
   Thinking, maybe we can do something like
   Model ID : ◀ model_name ▶

now those arrows left or right should open the model picker again for selection.

[abdulsaheel]

Makes sense @Revantark, made the change in accd8f6 . Once the server's model list is detected, Model ID shows as ◀ model ▶ and Left/Right reopen the picker; Esc keeps the current pick.

[Revantark]

Thanks.