Bayesian classifier cann't distinguish "SQL" vs "PLpgSQL"

[bzz]

Part of the #155.

After update to latest samples in #189, Bayesian classifier test fail to distinguish "SQL" vs "PLpgSQL" based only on content. Classifier weights are different in enry/linguist for the same document https://github.com/src-d/enry/pull/189#issuecomment-457559748

This most probably this has to do with with difference between tokenizations between two projects that going to be addressed in #193