Add an option to stream AI responses as they are generated to reduce downtime.

This allows users to read along and quickly decide whether to use the generated option(s) or try again.