`readability.main()` now depaginates using a recursive method.

This may or may not work for articles with many pages.

Otherwise Python chokes on the UTF-8 characters in the new regexes.

`lxml.etree.XPath` was not imported properly.

* Eliminate numerous careless mistakes.
* Fix a silly, random typo from a previous commit.
* Make the depaginator recursive like it should be.

Depagination:
* Activate only when new CLI option `-p` is given.
* Remove several lines of dead code.
* Restrict the list of possible links to the next page
  such that the URLs must contain as their protocol field
  either http or https.

Full HTML document (i.e. non-fragment) output:
* Include a meta-tag encoding declaration.
* Include the shortened title in html>head>title.

Previously, any URLs that were not protocol-absolute
would be rejected. Now, only the trivial pathology of
the "javascript" URL protocol is rejected.

Note that other pathological cases may exist.

Should infinite recursion be a sign that some pages are faulty?
Which ones?

For the record, the website that triggered the need for restrictions
on depagination was http://www.lifehacker.com since that site now
has a link whose text reads "Learn More »"
[they had a redesign to which that text refers]; "»" is the
"Right Angle QUOte" that is present in one of the regexes.

* Support non-UTF8 encodings by actually passing the `encoding`
  parameter from `main()` to `summary()`.
* Compose depaginated articles together properly and intelligently.
* Give the recursive depagination function a more proper name.