Refactor PreviouslyUsedKeyword assessment to include morphology

[nataliashitova]

While keeping the current implementation, add a second check where the keyphrases are being compared based on the base forms of all content words in the keyphrase. The base form will be (for now) guessed for English, for all other languages this new check will allow

• free order of words in the keyphrase (i.e., computer screen and screen computer will be matched as the same keyphrase for this assessment)
• non-reliance on exact match of function words (i.e., screen of the computer and screen of a computer will be matched) > only for languages with the function word list ready.

[hansjovis]

Currently devising a strategy how we can add morphology as described in the issue.
(And providing myself with an overview of how the previouslyUsedKeywords assessment currently works)

[hansjovis]

I would propose a strategy like this:

The function keyword_usage in class_wpseo_meta is responsible for querying the WordPress DB for posts with the given keyword. I would suggest to:

1. Give this function an array of content keyword base forms as an extra argument, on top of the single key phrase.
   E.g. apples and oranges => [ 'apple', 'orange' ].
2. We can get the base forms by applying a stemming algorithm. Many Natural Language Processing systems use the Porter stemmer for English.

I can think of two ways to adapt the retrieval of matching keywords:

Change the query
I have not found a way to do this yet, without also needing to adapt how keywords are stored in the DB. SQL is not great for soft-matching text strings. Using an AND of LIKE operators is the best candidate, but using this would e.g. match pineapple and oranges when querying on the [ apple, orange ] keywords.

Change the DB, and the query
This would be our best guess, in my opinion. However, it might prove difficult depending on how hard it is to migrate add an existing column with the new format in the DB. (add, since the previous way of matching should still work).