Improved paragraph algorithm

[jenniferjiangkells]

This PR implements improved paragraph detection algorithm.

1. Implements refine option in .process method of Note. This is configurable through refine_paragraphs option in AnnotatorConfig. The refine option merges any consecutive prose sections into one paragraph and merges a header with empty body with the next prose section. Address point below:

Header + blank line: if a paragraph has header and no body and no concepts detected, and the following paragraph is prose (i.e. does not match a header type), the paragraph type should apply to the following paragraph as well. e.g.

Problems:

- Aortic stenosis
- Diabetes
- IHD 

2. Implements .filter_concepts_in_numbered_list() in Annotator. This pipeline component (list_cleaner) filters out extraneous concepts in a numbered list and only returns the first concept detected in a line after a numbered list item. Addresses point below (in this example only CCF, IHD, Gallstones would be returned):

Numbered list: users may mark entries with numbers with varying paragraph sizes and location of line breaks. If detected as a numbered list then use only numbered items, keep only the first concept per numbered item. e.g.

Problems:

1. CCF -
- had echo on 15/6
- on diuretics

- awaiting pacemaker

2). IHD
3. Diabetes type 2

HbA1c = 78mmol/L
4 Gallstones

3. Makes loading of paragraph regex configurable with the rest of the lookups.