Before this, links would only be scraped from paragraphs and
rendered as a block after parent paragraph. This replaces this logic
with a generic links extractor that would recursively collect every link
from any parent node, including footnotes, blockquotes, and lists.
The renderer splits links block in three separate blocks, one
containing footnotes, the second containing images, and the third
links. Blocks are separated by a single line break.
This also makes the renderer skip link-only lists from rendering,
making them behave like links-only paragraphs (which are already
rendered as a series of consecutive links). The old behavior of
extracting links from links-only paragraphs (where the paragraph
itself is treated as just a set of links) remains, and can be still used.
The text extraction is also now unified across headings, paragraphs,
table headings, blockquotes, and other elements where it was used.
This fixes a number of bugs (like panicking on links inside headings)
and shortcomings where renderer would render text differently
based on container element type.
Fixes#17 and #23.
bcc3d8b730