Make the print method a bit more informative about the dictionary

[mattsq]

Currently we print the first 5 words by column position - which obviously results in word, word, word, word, NA when there's less than five words, which isn't great. But also it's not necessarily informative. Something more like the top 5 by class might be better:

"The top 5 words by class are:
Class %class1 : aaa, aba, bba
Class %class2: bba, bbb, aaa
..."

This'll help eyeball whether the representation actually achieves good class separation or not.

[mattsq]

Added a few commits to at least fix the NA problem, and to return the words sorted. Not sure we actually want to do this by-class one (maybe for a plot method?) but i'll leave this here for the moment.