Streaming output for language models

Replicate's API now supports server-sent event streams for language models, allowing users to receive real-time token-by-token output by setting the stream option to true and connecting to a stream URL.