Add an option to create or expand a TagDictionary with training data

It would be useful if we could expand or create the TagDictionary while training a POS Tagger model.

I propose that we add a new command line argument, -tagDictCutoff, that would trigger the creation / expansion of the dictionary. The cutoff would represent the minimun number of occurrences that a word tag pair would occur in the training data before it is added to the dictionary.

Further information can be found on this conversation: http://mail-archives.apache.org/mod_mbox/opennlp-dev/201205.mbox/%3CCA%2BiWThJNQzLSc3NmDLbEzaORDWnFgbk_id3SJjuELVRSoMTJzQ%40mail.gmail.com%3E