Moving this comment over from #13, as that issue will be closed soon:
When we add sparse segmentation data, we should consider varying write_empty_chunks
For segmentations with many zero values, this could have a large impact on write speeds / the final file size.