fix: use native iterable to read XLSX rows
We noticed that huge xlsx were very very slow, and it seems related
to the way we consume the rows since last refactor. Trying to call
one row at a time from its row number is slower than using native
rows
iterator.