What changed

• The parser now streams data through a 4MB read buffer.

• Failed rows are isolated instead of aborting the import. Each error includes the line number and reason, replacing the generic error message.

Supported encodings

• UTF-8: the default, detected via BOM and heuristics.

• UTF-16 LE/BE: detected via BOM, with surrogate-pair support.

• Windows-1251: detected heuristically for legacy Cyrillic exports.