Standardise CSV reading

[vreuter]

The assumed CSV structure for locus-specific points differs from the assumed CSV structure for nuclei centers/labels.

For locus-specific points, we assume:

• no header
• no index column
• 3D (z, y, x)

For nuclei centers/labels, we assume:

• header
• index column
• 2D (y, x)

We should standardise parsing of CSVs, and more generally abstract/generalise to parsing "records" which may come from CSV, .roi, JSON, etc.