Ebookmaker

- fix reversion in txt production because GutenbergTextParser needs caching

- caching the html tree in the parserwas causing the epub build to interfere with the epub3 build, so a reset method has been added for parsers to make themselves safe for reuse. Raw bytes are still cached, since most of the benefit of caching is reading from disk.

- allow figure in body
- let PDFWriter create a directory

- fixed bad bug finding footer in texts generated from TEI

- restore stylesheet in cover wrapper
- PG producers used to do all sorts of crazy things in css comments, such as nesting CDATA sections: `/*<![CDATA[ */`. tidy used to clean this up for us. Now we're removing css comments entirely.
- what's more, the css sometimes contained malformed xml comments

- fixed parsing no-footer text files (stupid mistake!)
- fixed an issue where css files disappear because they are empty but there are still links to them
- added some file checks and a custom exception so that it's clearer what has happened if you give ebookmaker a bad file.
- removed a css link in wrapper that was was getting stripped, then re-inserted later.
- ancient browsers didn't understand stylesheets, so xml comments were used to hide the style text. Our CSS parser is too modern to remember this, it seems. So we needed to un-comment style text. Probably was another thing that tidy was doing without telling us.