Add validate_chunk flag (reorganized and updated)

[lylyhan]

A new flag validate_chunk (input to SpeakerDiarization class, defaults to False) will determine if chunk is skipped when exceeding max_speakers_per_chunk
3 changes if validate_chunk=True :

1. SpeakerDiarization.prepare_chunk can return None
2. SegmentationTask.train__iter__helper will skip current chunk if exceeding max number of speakers
3. SegmentationTask.prepare_validation will skip current chunk if exceeding max number of speakers. If none of the validation data satisfies criteria, it will return the full "unvalidated" validation set.