* No count limit was being passed in the calls to replace() in
   core.util.ZipStreamGenerator.  The replace was intended to strip
   the base folder from its childrens' paths.  Instead it could mutate
   the intended file names if the base path string appeared multiple
   times.  Eg. Try to zip a folder called `foo` that contains a
   subfolder called `subfoo/` and a file `subfoo/bar.txt`.  You will
   end up with a top-level file called `subbar.txt`, because
   `'foo/subfoo/bar.txt'.replace('foo/', '')` will strip both
   instances of `foo/`.

 * Passing 'foo/bar/' to bitbucket's url-building methods results in
   the slash being percent-encoded ('foo%2Fbar'). path_tuple() will
   return the path as a tuple, without the leading empty root
   string: ('foo', 'bar', )  The url builders will properly encode
   this as 'foo/bar'.

 * For recursive operations like DAZ, self._parent_dir is not
   guaranteed to be the path's parent (though it should be an
   ancestor).  Test whether that assumption holds true, and if not,
   fetch the parent dir listing.

 * Bitbucket allows all characters except / in filenames. Bitbucket
   API urls have the following requirements for character encoding:

   * must encode: \#%?
   * optional encode: [{]}|"<>~`!@ $^&*()+=,;:'
   * encoding not used: ._-

   Support for the percent sign in file names is spotty right now, b/c
   the version of furl we use to build urls is buggy wrt %:

   ````
   In [12]: furl.furl('http://foo.com/m%o/').url
   Out[12]: 'http://foo.com/m%o/'

   In [13]: furl.furl('http://foo.com/m%25o/').url
   Out[13]: 'http://foo.com/m%o/'

   In [14]: furl.furl('http://foo.com/m%2525o/').url
   Out[14]: 'http://foo.com/m%o/'

   In [32]: furl.furl('http://foo.com/a%23%3F%5Ce/?bar=baz').url
   Out[32]: 'http://foo.com/a%23%3F%5Ce/?bar=baz'

   In [33]: furl.furl('http://foo.com/a%23%3F%5Ce%25/?bar=baz').url
   Out[33]: 'http://foo.com/a#?\\e%/?bar=baz'
   ````