One increment of a AiProvider.completeStream reply.
The stream ended in failure — before, or after, some Tokens already emitted.
A piece of the model's reply text, in generation order.