Use wider/different CSV input table for benchmarking

[matt]

As of now, the generated test file (/tmp/hello.csv) contains a series of rows containing hello,","," ",world,"!". However, this has a very small number of columns per row, small average number of bytes per cell, and high average number of quoted values per row-- all of which can significantly impact parser speed-- whereas many (I’d guess, the vast majority of) real-world CSV use cases have very different profiles. If we submit a pull request, would you support the addition of another input file for benchmarking purposes? For example, https://github.com/petewarden/dstkdata/blob/master/worldcitiespop.csv