Patient CSV cannot name the DNA and RNA arms of one specimen

🤖 Written by Claude.

Split out of the TSO 500 plan (SACGF/variantgrid_sapath#431), a leftover from #1704.

The patient CSV import's columns are all `SPECIMEN_*`, so it round-trips one `Extraction` per `Specimen` and cannot name the DNA and RNA arms of one block (TSO 500's `2600000001C` and `2600000001B` are two extractions off one specimen). The API (#1707) can express this; the spreadsheet cannot.

Worth doing when a consumer needs it — nothing currently does. When it is done, note `Extraction.nucleic_acid_source` and `Extraction.extraction_date` are what tell two extractions of one specimen apart, and `Extraction.reference_id` is unique per specimen rather than globally.

Start with the patient CSV import and compare its specimen-based columns with the representation in API #1707. Use Extraction.nucleic_acid_source and extraction_date to distinguish DNA and RNA extractions while preserving reference_id semantics; done means one specimen's multiple extractions can round-trip through the spreadsheet.

Written by the indexing model from the issue text.