List Admin Llm Provider Models

Return models for provider plus structured failure context.

Enumerates settings.llm_api_base, matching what the agent loop passes to amessages on every call. Previously this called any-llm with no api_base at all, so the listing went straight to the provider's own API while the agent talked to a gateway. On a gateway deployment that failed outright, because the gateway's virtual key got presented to the real provider ("401 invalid x-api-key"), and an admin could never see the models they were actually able to call.

Deliberately takes no caller-supplied api_base. The OSS sibling (/api/user/providers/{provider}/models) accepts one because its settings form passes it, but the admin UI does not, so the parameter would be a curl-only surface whose only real effect is to let an admin make the server deliver a provider API key from its environment to an arbitrary host. If the admin form ever needs to preview a candidate endpoint before saving it, add the parameter together with URL validation, not before.

Never raises 4xx/5xx for "this provider cannot list models" or "the provider's API call failed". Those are normal admin states; the UI needs to render them, not see a generic 502. We only let through framework-level errors (e.g. validation), which FastAPI handles.