Per sentence (or regex customizable) split for TTS

[hat-commits]

Is your feature request related to a problem? Please describe.
TTS requests asks for way too much at once, leading to long time until first response.

Describe the solution you'd like
Per sentence TTS request split that are queued and requested as they go along, to minimize both the time until first utterance and reduce waiting.

[rasmus-z]

I imlpemented that in my fork and here's a pull request if you are interested: #78