Adding .__iter__() is not strictly required to enable iteration, but it
improves the performance of iteration significantly by avoiding the
default implementation which would repeatedly parse the bookmark pairs
in the document.

* Sequence is now in `collections.abc` and will be removed from
  `collections` in 3.8.
* zip() returns an iterator in Python 3 and must be wrapped with
  something like list() or tuple() to be realized.
* Remove support for 2.6.
* Add tox tests for 3.6 and 3.7.
* Update .travis.yml to match tox versions.

Raised the bookmarks to the paragraph level

Implemented possibility to remove bookmarks

Feature/bookmarks

- Multiple open bookmarkStart elements resulted in identical bookmarkStart ids.
- Modernized oxml __init__ formatting for future developments
- Added Paragraph.start_bookmark() and Paragraph.end_bookmark() methods to support paragraph level bookmarks.