I am inserting close to 2 million rows into three different tables using bulk load. One of them represents 1.2 million rows with 7 columns.
I had a look to heaptrack's collected data, and it appears that into_rows produces many Vec grows :

Indeed, the helper functions all use ::new instead of ::new_with_capacity. If I initialize them with the correct size, TokenRow::push is no longer present in the report and the allocation count decreases.