Ebookmaker

- fixed minor issue where the spider was getting confused when iterating on WrapperParsers.

- handle invalid quantization table in jpeg files when using quality 'keep'
- respect rel="nofollow" attribute: this allows authors to link to an alternate version file in html without duplicating content in the EPUB file.
- set wrappers to nonlinear in spine.
- fixed bugs and ugliness in toc generation
- when the same header level is consecutive, only one toc item is generated. (We see this used to make multiline heads or titles)
- toc normalization made a hash of the toc. Now the toc is normalized in the epubwriter, not in the parser.
- add display:block to standard css sheet to prevent hidden headings from breaking kindlegen
- added a configuration option to use calibre (or whatever!) for non-supported in kindleget languages.

- Fix reversion in 0.9.2 which caused CoverWriter in EbookConverter to fail. (It uses ImageParser to convert images from PNG to JPEG
- add a test to check this

- note that EbookMaker is no longer installable in Python 2.7 (thanks cpeel)
- clean up pipfile and gitignore pipfile.lock (thanks cpeel)
- fixed bug where filepaths need escape in HTML
- fixed issue where compression was expanding compressed jpegs (thanks choward)
- adde EBOK flag in MOBI when using ebook-convert (thanks rfrank)

Minor bug fix and optimization.
- picsdir builds weren't recognizing when it was copying files to themselves
- "broken" images are now inserted when an image is missing
- title attribute in wrappers needed escaping
- build times are now reported to logs
- small optimization with preparse on image parsers

- linked images (image files as targets of links) are now wrapped in html, fixing display in ADE.
- linked images are compressed to 1MB if possible (changed from 128K)
- inline images are compressed to 256K if possible (changed from 128K)
- all images are limited to 5000x5000 pixels (was 800 x 1280)
- PNG images are scaled to meet the image filesize targets (previously no scaling of PNG images)
- L format JPEG images (greyscale) are no longer converter to RBG (thanks rfrank)
- generated covers are now 1200 x 1800
- covers and "important" images in noimages builds are scaled to 64K (previously no scaling)

General bug fixes
- eliminated double parsing due to first pass using raw paths
- when tidy fails, the (huge) error trace is only logged once.
- generated html is no longer overwritten by empty results