CsvExporter renders a null struct row as {"field":null,...} instead of an empty cell

[dfa1]

Summary

A Struct-dtype column where the entire row is null (not just individual fields) renders as a JSON object with every field null (e.g. {"name":null,"count":null}) instead of an empty CSV cell, unlike every other nullable column type in CsvExporter.

Discovered while validating the fix for #268 (chunked List columns) against the Raincloud conformance corpus: oasst1 has a nullable struct column (labels) where some rows are entirely null. After #268's fix the vortex-java export no longer crashes, but its output diverges from the parquet oracle on those rows.

Root cause

StructEncodingDecoder (reader/src/main/java/io/github/dfa1/vortex/reader/decode/StructEncodingDecoder.java, ~line 55) represents struct-row nullability by wrapping each field in the same MaskedArray validity, not by wrapping the whole StructArray once:

return structValidity != null ? new MaskedArray(field, structValidity) : field;

So a null struct row decodes as a StructArray whose fields are each MaskedArray-wrapped-invalid — the StructArray itself is never masked.

CsvExporter.cellValue (csv/src/main/java/io/github/dfa1/vortex/csv/CsvExporter.java, ~line 165-196) only renders "" for a null value when it sees a MaskedArray wrapping the whole column value:

case MaskedArray ma -> ma.isValid(rowIdx) ? cellValue(ma.inner(), rowIdx) : "";
...
case StructArray sa -> jsonObject(sa, rowIdx);

A StructArray at the top of the switch always goes straight to jsonObject, which calls jsonValue per field — each field is individually masked invalid, so jsonValue's own MaskedArray arm renders "null" for each one, producing {"field1":null,"field2":null,...} instead of checking whether every field is invalid for this row (meaning the row itself is a null struct) and rendering "" instead, matching the parquet oracle's isNull(fieldIndex) → empty-cell behavior for a null top-level field.

Expected

cellValue's StructArray case (and/or jsonValue's, for a nested null struct inside a list/struct) should treat a struct row where every field is invalid as the row itself being null, and render an empty cell / JSON null respectively — matching every other nullable-column type in the same file.