Stream unknown-length input with continue=true and identical fields, add your own spaces between chunks, and finish with an empty transcript and continue=false so the tail gets generated.

Context: Official docs (Cartesia working with websockets): when streaming text of unknown length, keep sending chunks with continue=true and keep every other field identical across chunks, including voice and model settings. Supply spaces explicitly between chunks or words run together. To finish, send an empty transcript with continue=false. If you just close the socket without the empty-transcript message, the trailing audio may never be generated.