fromConfiguration

What a model is for when the runner itself does not say.

Docker's /v1/models and Ollama's /api/tags list names and nothing else, so configuration decides: a model some embedding-services entry or embedding role names is an embedding model, and anything else is a chat model.

ONE function rather than one per runner, because both runners' startup registration applies this same rule and the guarantee that matters - a model cannot land in one category at boot and the other when pulled later - is only as good as the two copies staying identical. LM Studio does not call this; it reports a type per model.