Make included benchmark pluggable for different dataset

[raganhan]

It would be nice to make this pluggable for different data sets as well, since the performance of the path extractor is highly dependent on the characteristics of the data. It would be good to eventually provide results for several data sets along with a description of the data and what was skipped. This doesn't block the initial release.

Originally posted by @tgregg in #8

[raganhan]

It's probably good to not force the text version in other benchmarks as it was included mostly to demonstrate you get equivalent performance to the DOM as there is no skip scanning