These functions, which I had to implement in Python 2.2, are now
available in standard Python.

Move CountingProbDist to learning.py and use it in NaiveBayesLearner. Make NaiveBayesLearner use target-value frequencies too.

 * Use weighted_sampler() for the sampling.
 * By not using a defaultdict, keep n_obs always up to date.
 * Don't inherit from ProbDist, at least for now, since we don't
   use anything from the superclass, and if we did it would break
   (e.g. __setitem__).
 * Remove __len__ since nobody uses it.
 * Use heapq.nlargest.

Also separate random doctests from others, and tweak docs.