Add latency info

[clemlesne]

An extremely important parameter for UX in realtime scenarios is first token latency with a certain amount of input tokens (ie 1k and 10k). It applies on STT, TTS and LLM. This would be a useful data.