The epub-builder still downloads the image, but all the html-mangling
is done in the extraction process now.
Turns footnotes into a chapter-object, for easier processing later on.
Specifically, added image_selector for arbitrary sites that allows
selecting img tags from chapters, downloading them
and embedding them within the resulting epub.
In the case of Pale, this means that the character banners and
extra materials do not require an internet connection to view.
Also made the two pale.json's more consistent (pale.json now correctly
includes the title of the chapters).
Makes a generic _clean function on Site that can be called. Will
probably want to migrate some other generic bits into there after
analysis of what's *really* generic.
Each `Chapter` object had a reference to the entire page tree, meaning that the program rose in RAM usage by a lot.
Transformed Worm to be with next_selector so the chapters are correctly ordered, E.2 is not skipped and the download does not crush due to `?share=twitter` url matched before.
Fixed Worm titles.
* next_selector: find next content page, if not using chapter selector
* content_title_selector: pull a chapter title from the content
* content_text_selector: pull specific text from the content element
`content_selector` will now fetch all content elements on the page, each
as a Chapter, not just the first one that matches.
