Better extraction & better http handling

• Requests used for images. Same http session is used for all requests.
• Analyze all possible text root nodes and select best one, do not stop on first text root node candidate
• Improve text selection filters

Move to requests as network library

Draft new release

1.0.28:

  * Move to requests as network library

Python 3 support

Enable python3 support

Fix unicode processing + ` ` support

• As STOP_WORDS are stored in unicode format we should keep our words candidates in unicode also to be able to compare candidates against dictionary correctly
• With some languages, short stopwords are linked to the next word in the sentance with no-breakable-space. To designate those stop words we should support nbsp when tokenizing. Russian is an example. So this fixes grangier#223