Added assistant screen power modes and inactive animation states

Power modes can now be controlled through the screen which also control animation states.

Tools were added for Text To Speech benchmarking.

All config values are centralized for a more organized functionality.

All used models during a session are now warmed up from start for faster responses.