Latency improvements with the new manual VAD algorithm

[clemlesne]

Done:

• Implement monitoring metrics to follow improvements

Todo:

• Parallelize more TTS and database calls (study OTEL traces for opportunity confirmation)
• Reduce dependency calls before sending call to the LLM or defer them
• Compress the prompt (LLMlingua?)
• Use a LLM with a lower latency (Phi 4?)
• Trace the code executions with local debugger to pin points unseen optimizations

[clemlesne]

OTEL gauges and counters has been added to follow technical and functional metrics.

Notably:

• call.aec.droped, number of times the echo cancellation dropped the voice completely.
• call.aec.missed, number of times the echo cancellation failed to remove the echo in time.
• call.answer.latency, time between the end of the user voice and the start of the bot voice.

See 59e0d68, 34ca384, a2b484c, 24144a9.