Add sectioner code

[marcverhagen]

Run this on a TarsqiDocument and have it add docelement tags.

Focus on some fairly generic heuristics for splitting and a few specific ones for the Thyme data. Generic heuristics to use:

• Short lines with only a few words, not ending in a period, potentially starting with a number or some other indicator that something is a header.

• Empty lines, almost always indicating that the text above and below are not in the same sentence, this is similar to what the current document structure parser in https://github.com/tarsqi/ttk does.

• Certain XML tags when available.

The Thyme data have many variations of the short line heuristic. There are enumerations:

Height=150.00 cm,
Weight=39.40 kg,
Height=59.06 [in_i],

Singer - Hospital Summary
Admission Date: 05-Dec-2005  Dismissal Date: 07-Dec-2005
Contributing Author: Kxixcaj Q. Oarvui

The Stanford splitter puts these in one sentence, and a bad one that explodes the parse, so we should at least not recognize this as one sentence.

Section headers are also a variation of the short line heuristic:

[end section id="20104"]

These are also usually separated by whitespace, but not always so adding the brackets to the heuristics for Thyme may be good.

Also, in Thyme, a couple of consecutive words in ALL CAPS, starting at the beginning of a line (even long lines) are a strong indication that this may be a section. This needs to be explored further.

[marcverhagen]

Some of this was done in commit 57e26cf, but no effort was made yet to split general heuristic and THYME heuristics.