Perfect Diarization

Enable perfect diarization for real-time and async transcripts to guarantee speaker attribution works, even when participants talk over each other.