Improve byline algorithm

[leibovic]

Still a WIP, as I'll have to rebase on top of #37

This fixes at least these two testcases:
http://www.salon.com/2015/01/02/coca_colas_anti_american_outsourcing_scheme_how_big_soda_gets_the_public_to_shoulder_its_costs/
http://alistapart.com/article/designing-for-easy-interaction

Using the author meta tag is definitely a strict improvement, as is checking the nodes before throwing them away. In order to get the salon.com testcase to work, I added logic for preferring a rel="author" element, but I don't know if that's too specific to that testcase.

Also, I don't like the var self = this; crap, but it was the quickest way to getting this to work to test :)

[leibovic]

I updated this to not favor the rel="author", since that was causing one of the tests to fail (a valid failure).

[leibovic]

With the new _removeAndGetNext logic, I realize this still has the problem of not checking the children of nodes that we're about to remove (e.g. digging through a header node to look for a byline). Perhaps we should try recursively looking through nodes before removing them if we haven't found a byline yet.

[gijsk]

With the new _removeAndGetNext logic, I realize this still has the problem of not checking the children of nodes that we're about to remove (e.g. digging through a header node to look for a byline). Perhaps we should try recursively looking through nodes before removing them if we haven't found a byline yet.

This sounds like a good idea for a followup, as discussed IRL.

[gijsk]

Please update the comment. :-)

[gijsk]

This generally looks great, just a few nits, for which, see my earlier comments. :-)