* feat: Add missing TTS models and voices
* feat: Add new instruction field to create speech request
- From docs: Control the voice of your generated audio with additional instructions. Does not work with tts-1 or tts-1-hd.
* fix: add canary-tts back to SpeechModel
