Recluse

..........................
* NON-BACKWARDS-COMPATIBLE change to open_with_unicode parameters: now compression_type is a keyword argument; when left out, transparently falls back to opening with no compression, and matches the interface of regular Python open.
* Cleaned up a lot redundancy in open_with_unicode (Thank you, Allison Kaptur!)

..........................
* Wrapped the shutil.rmtree in vocabulary_generator in case of errors due to timing effects of .nsf files.

..........................
* Refined the vocabulary_cutter and vocabulary_generator so they can optionally use a frequency cutoff instead of a size.

..........................
* Refined the nltk_based_segmenter_tokeniser punkt object option; the punkt_obj must be an sbd attribute of an nltk.tokenise.punkt.PunktSentenceTokenizer.

..........................
* nltk_based_segmenter_tokeniser can now read in a punkt object from prior work, for example, a pickled one.

..........................
* Made a new module article_selector, intended to replace article_randomiser.