Tacl

* Fixed bug that meant extended diff results were immediately
removed after being generated. (44)

* Noted on the tacl report help the reason for --extend coming
before --reduce when both options are specified.

* Added automatic removal of filler n-grams from diff results. Only
n-grams that are entirely composed of (n-1)-grams, or which do not
overlap with any (n-1)-grams, are kept. This then allows for
reduce and extend to work correctly on diff results.

As a result of this change, the API for all difference queries has
changed.

* Add example invocations of tacl subcommands. (30)

* Added basic formatting of highlighted text to match that of the
stripped text.

* Changed calls to pandas' to_csv method to keep up with current API
(use 'columns' instead of 'cols').

* Changed call to pandas' sort_index to sort_values to keep up with
current API.

* Specified minimum version of pandas (0.17.0) in setup.py.

* Changed arguments to sdiff and sintersect subcommands, to work
around Python bug https://bugs.python.org/issue9338

* Changed tacl highlight and align to use Jinja2 templates. This
paves the way for the HTML/CSS/JS-rich output of future scripts.

* Added zero-fill option to tacl report, supplying result rows with
a count of zero to witnesses of a text that has at least one
witness with a non-zero count.

* Added collapse-witnesses subcommand to tacl-helper, to collapse
result rows for witnesses to the same text with the same count
into a single row.

* Fixed a bug in tacl report's extend if the same witness is
associated with more than one label in the results.

* Added check for texts being relabelled in a catalogue file.

* Added check for only a single label being supplied to a query.

* Added support for multiple witnesses to a text. Witnesses are
automatically extracted from the CBETA texts, and results now
include a 'siglum' field indicating which witness of the text a
result comes from.

* Added total count of n-grams occurrences per text to 'tacl search'
results.

* Split 'tacl strip' into 'tacl prepare' and 'tacl strip'.

* Reimplemented supplied queries to be more useful and not have
hidden gotchas.

* Fixed a bug causing 'tacl stats' to sometimes return incorrect
results.

* Added a tacl align command, to output aligned sequences of
intersections between all pairs of differently labelled texts in a
set of results.

* Added LICENCE and CHANGES to MANIFEST.in.