#84 WIP convert to binary instead of to utf8

[jssuttles]

trafficstars

I tested this with the file that I gave, and it seemed to work.

[pieper]

Thanks for starting this. Since neither of us is sure if this is the overall correct answer I'm going to put this on the back burner. To move this forward I think we need to cross-check with other implementations on a variety of datasets as discussed in #84.