benchmark: compare the performance trade-offs of validating each `.csv` row on read, vs. validating only the final constructed `.csv`

[jdhoffa]

          I'm happy to merge this as-is, but it would be interesting to see what the performance trade-offs are with validating each row on read. 

Something to look into in the future

Originally posted by @jdhoffa in #23 (comment)