- Add missing runtime dependency on idna

- version update to 19.0.0
* When passed to `str()` URLs now stringify to usable URL
strings. (See #49)
* Switched off of Python's built-in IDNA facilities to using the
[idna](https://pypi.python.org/pypi/idna) package. Not only is it
much more modern, it's also much more strict and correct in its
output. (See #19 and #56)
* Added new `DecodedURL` type with almost-identical API to the normal
`URL`, except that it automatically handles reserved characters in
argument values passed to its methods. (See #6, #11, #44)
* Added top-level `parse()` convenience function that now represents
the main entrypoint to hyperlink.
* Accept dictionaries as ‘query=’ arguments, in addition to
sequences of tuples (see #50)
* `URL.child()` will no longer fail when child gets no segments (#42 and #43)
* `URL.normalize()` now supports encoding stray/unmatched `%` characters
in character-encoded fields (userinfo, path, query string, fragment)
(see #61, #62)