…mment

…r rewrite

… compatability!

Prior to this change, the public API of the PorterStemmer was a mess. NLTK's version was
based off Vivake Gupta's implementation at http://tartarus.org/~martin/PorterStemmer/python.txt,
endorsed by Martin himself at http://tartarus.org/~martin/PorterStemmer/. However,
Gupta's implementation is a shoddy port of Martin Porter's own implementation in C,
and had several vestigial quirks lying around. These include the claim that the stem()
method takes a "char pointer" as an argument (no such thing in Python) and the need
to pass in start and end indexes between which stem() should read the word from the
given char array.

At some point in nltk's history, during or prior to the 2006 commit that added porter.py
to the current Git repository:

    edf4677

this was "solved" by renaming Vivake's stem() method to stem_word() and creating
a wrapper for it called stem() that conformed to the StemmerI interface.

This was completely pointless; the right thing to do would've been to remove the
unnecessary parts of Vivake's stem() method and thereby acheive conformity to StemmerI.
This commit does this, but at the cost of breaking backwards compatibility for
anyone who was using stem_word(word) instead of stem(word); those people will
need to adjust their application code when updating to the latest version of NLTK.

I don't even know where this came from. It's not part of the Vivake Gupta version
at http://tartarus.org/~martin/PorterStemmer/python.txt

Regardless, there's no reason for it to remain here.

…d Snowball

Handling of upper and lower case is not specified in Martin Porter's
"An algorithm for suffix stripping" paper; the algorithm description
there never even mentions the existence of difference letter cases.
Nor does Martin's C implementation of the stemmer at:

   http://tartarus.org/~martin/PorterStemmer/c.txt

handle case in the way that NLTK's version has been doing; instead,
it simply requires that the user convert their word to lowercase
before calling stem().

Since there is no Porter-specific reason to preserve our (odd)
behaviour here, and our other StemmerI implementations don't do it,
we should probably purge it, as this commit does.

- Making the code more readable and including quotes in comments that allow the code
  to be matched up with Porter's published algorithm
- Marking NLTK-specific departure points properly

…mpliance

… the algorithm. Also sort out the docs.

…used in the AUTHORS file