CSV header-mode validity check must handle object rows

Problem

CsvPlugin is supposed to deserialize plain CSV into a Slate table.

Valid input like name,age\nAda,36 returned undefined on the default path because Papa Parse was configured with header: true, but the validity check still treated the parsed rows like arrays.

Root cause

deserializeCsv used one validation shape for both parse modes:

data.length < 2 || data[0].length < 2 || data[1].length < 2

That only makes sense for array rows. In header mode, Papa returns objects plus meta.fields, so:

data.length counts only data rows, not the header row
data[0].length is undefined

So valid header-based CSV was rejected before the AST builder ever ran.

Fix

Split validation by parse shape:

header mode uses meta.fields for column count and requires at least one data row
array mode keeps the old row-length checks
error tolerance still compares parser errors against parsed row count

That makes the default plugin path valid again without weakening the malformed-CSV guardrails.

Verification

These checks passed:

bash

bun test packages/csv/src
pnpm test:slowest -- --top 20 packages/csv/src
pnpm turbo build --filter=./packages/csv
pnpm turbo typecheck --filter=./packages/csv

Prevention

When a parser supports both header mode and array mode, test both. They are not the same data shape, and pretending otherwise is how you ship a broken default.