CSV Import With Per-Row Errors

Specify a CSV import that validates every row and returns a file of errors instead of failing the whole upload on row 2. IMPORT Entity: [USERS | PRODUCTS | TRANSACTIONS] Columns required: [LIST] Columns optional: [LIST] Unique key: [EMAIL | SKU | EXTERNAL ID] Max rows: [N] What a duplicate does: [SKIP | UPDATE | ERROR] DELIVER 1) The parse step: encoding, delimiter sniff limits, header normalization, and the reject if the file is not CSV. 2) A Zod or similar schema per row, plus cross-row checks (duplicate keys inside the file). 3) The result object: inserted, updated, skipped, and an error list with row number, column, and a message a human can fix. 4) An error CSV the user can download, same columns plus an error column. 5) Transaction behavior: all-or-nothing versus partial commit. Pick one for this entity and say why. Show how a partial commit stays idempotent if they re-upload the error file. 6) Timeout: this runs as a job if over [N] rows. Say how the UI polls status. Do not eval cell contents. Do not import formulas as values without stripping a leading equals sign, and explain that spreadsheet injection risk.