Add naive Bayes classifier for text

[Erik Rose]

We'll probably need one to distinguish Project Smoot's Article class from its Techie Article class. My initial musings have the classifier represented as a scoring callback:

rule(..., score(bayes(...)))

It, like any scoring callback, would take a fnode and probably operate on its innerText. Also, like any scoring callback, it would return a value in 0..1, which the rest of Fathom could threshold or mix into other types, as usual. An open question is how to train it.