Group CSV errors by repair action, not row number

CSV import errors should group identity, required field, format, and ignored column problems.

Raw CSV import errors often tell users where a problem is, but not what to do next. Grouping errors by repair action is clearer: identity errors, required field gaps, format repair, and ignored columns. Then show counts and examples instead of dumping hundreds of row-level failures. This also helps support because the user can say “I have 18 duplicate identity rows” instead of “the importer failed.”