Support explicit start/end timestamps of speech activity detection (VAD) given by the user + add vad segments in the output

[Jeronymous]

• option vad can be a list of explicit timestamps (ex: [(0.0, 3.50), (32.43, 36.43)])
• Transcription results include a new key "speech_activity" with the timestamps of the VAD

  "speech_activity": [
    {
      "start": 0.0,
      "end": 3.76
    },
    {
      "start": 32.37,
      "end": 36.43
    }
  ]