fixed local tfidf to compute (doc, ngram) similarities and NOT cumulated scores per ngram (TODO: discuss parallel changes for global tfidf)

NLP stuff: text cleaning after file parsing and after ngram extraction, more stats about maplist, details for postaggers, params for max ngram length and lowercase conversion

[FACTO] listes (map, stop) + scores (occurrences, specificity) + ngrams tools to insert ngrams fastly. Need to be tested after NodeHyperdata table creation.

journals view, with new API route for journal facet, dynatable css factorized (applies for all corpora templates) and menu.html providing a corpora navbar to switch between journals and documents