Improve diff calculation

[RouxRC]

• [x] chiffres négatifs in tables https://www.lafabriquedelaloi.fr/articles.html?loi=pjl14-596
• [x] majuscules accents ? https://www.lafabriquedelaloi.fr/articles.html?loi=pjl07-405&article=1er&etape=15 https://www.lafabriquedelaloi.fr/articles.html?loi=pjl17-119&article=10&etape=9
• [x] diff case https://www.lafabriquedelaloi.fr/articles.html?loi=pjl13-215&article=1er%20bis&etape=15

[mdamien]

Also, we could improve the speed (currently it's one of the main performance bottleneck):

• ~using PyPy instead of CPython (20x faster)~
• ~using the python port of diff-match-patch~ (done)
• bonus: running multiple diffs/projects in parallel (since the diff process max out only one CPU)
• use levenshtein for similarity distance (100x faster)

[boogheta]

We could try with final texts to not take account of the final law number of the text added within such as "loi XXX du XXX" which create diffs