Track html tags

[jelmervdl]

Implementation of the html tag per paragraph part discussed in #40.

Split off from #35 so that can be merged while figuring out how to do this can continue..

[jelmervdl]

I've started adding tests for html parsing, and it is showing the limits of our html parser.

We don't keep a stack, so for example, this causes a problem:

<div>
  <p>Text</p>
  not block text
  <p>Paragraph</p>
  final line
</div>

Tags are assigned once the opening or closing element of a block element is scanned.

Here, not block text is also assigned a <p> tag because there is no </div> there, and the scanner's tag_name is not reset to the enclosing element (div) once </p> is scanned. It is still set to p after scanning </p>. The final line bit will be correctly assigned </div> because tag_name is once again set to div since it just scanned a closing element.