PREFER_ON_DEVICE

Prefer on-device inference, but fallback to in-cloud if unavailable.

In this mode, the SDK will attempt to use on-device inference first and, if the on-device model is unable to generate an answer, it will retry using the cloud model.