Meta tag extraction fixes

[karls]

I discovered that extracting information from the HTML was a bit broken -- at least I couldn't get it to work properly. The article.meta_data dict contained the following after parsing:

defaultdict(<type 'dict'>,
{'CPS_SITE_NAME': {},
 'application-name': {},
 'Headline': {},
 'IFS_URL': {},
 'Section': {},
 'UKFS_URL': {},
 'CPS_AUDIENCE': {},
 'msapplication-task': {},
 'Description': {},
 'CPS_ID': {},
 'CPS_PLATFORM': {},
 'msapplication-TileColor': {},
 'contentFlavor': {},
 'msapplication-window': {},
 'CPS_SECTION_PATH': {},
 'twitter': {'card': 'summary'},
 'bbcsearch_noindex': {},
 'msapplication-tooltip': {},
 'CPS_ASSET_TYPE': {},
 'msapplication-TileImage': {},
 'msapplication-starturl': {},
 'viewport': {},
 'OriginalPublicationDate': {},
 'og': {'url': 'http://www.bbc.co.uk/news/world-africa-16377824',
        'site_name': 'BBC News',
        'image': 'http://news.bbcimg.co.uk/media/images/57644000/jpg/_57644047_armed-lou-nuer-youth-in-lik.jpg',
        'type': 'article',
        'title': "'More troops' to South Sudan town"},
 'THUMBNAIL_URL': {}})

Only og, twitter etc keys were parsed correctly. This patch fixes this bug and all the keys are correctly parsed.

[codelucas]

This has been merged, nicely done.
Btw, check out my newest commit. It's a refactor on the meta data
extraction and also a more comprehensive test case.