feat(x spaces): Don't wait for mute, wait for silence (elizaOS#2576)

* use getSetting to get the api key

* use debounce for transcription processing instead of unmute

* prevent transcription while previous transcription is processing

* rename

* rename

* make sure the flag would be reset

* clean all buffers when starting to processing audio

* use one timeout instead so only last user audio would be handled

* implement interruption

* process reply message by context and state

* create template file

* rename

* add should response

* use constant

* clean code

* replace console log

* add missing type

* add optional for twitterSpaces