Make cascade of different content extractors explicit and configurable

[adbar]

So far Trafilatura is entwined with a version of readability-lxml, it also uses jusText as fallback before triggering the baseline extraction as last resort. This combination is robust and performs well in the benchmark, however it can be beneficial to refactor the code so as to expose the extractor chain.

The current configuration can be written as follows:

• fast mode: ["trafilatura", "baseline"]
• normal mode: ["trafilatura+readability", "justext", "baseline"]