Update diarize.py

[ConProgramming]

• Now assign_word_speakers also returns chunk level segments split by new speaker
• Now assign_word_speakers also returns a set of unique speakers

[ConProgramming]

I was also going to change word_segments to have "speaker": speaker and "text": seg["text"][int(wrow["segment-text-start"]):int(wrow["segment-text-end"])]...

But according to the code comment the .srt format depends on it being "text": f"[{speaker}]: " + seg["text"][int(wrow["segment-text-start"]):int(wrow["segment-text-end"])]?