Changes to reflect all modifications discussed in nltk#924

Warn a user when there is no ROOT node, instead of throwing an exception...

Add MT Visualization

* arguments are in same order as re.search (pattern then objects being searched)
* dropped support for searching a single tree; API requires a sequence
* return an iterator instead of a list
* use sphinx docstrings

tgrep: tgrep_positions and tgrep_nodes can take lists of trees

  where split(' ') attempts to read lines from n-gram files.

a couple of bug fixes in dependencygraph.py

Fix bug #963 finding jar files

1. `networkx.XDigraph()` was long ago [renamed](http://networkx.lanl.gov/reference/api_0.99.html) to `MultiDiGraph`
2. `cycle_finding_demo()` was out of date

* Python 3 compatibility changes to TextCat module: max integer
* Multiple changes in both corpus reader and TextCat module to
  make language detection work on unicode instead of bytestrings
* Removed own version of trigrams in TextCat module, now using
  one from nltk.util

This should resolve issues faced at #944. However, there is code that
depends on a fake root node, for example the tree visualisation code reads this and FStructure.to_depgraph() sets it.

Add support for Google App Engine

Tgrep module documentation

Update filenames and locations to track changes in Stanford tools; change names POSTagger->StanfordPOSTagger and NERTagger to StanfordNERTagger; fix package level imports to include the taggers

Merge nltk_tgrep into nltk (#931)

Fix bug shown in jenkin server ...