Abstract formats and restructure the repo

[Pierlou]

The idea is to get rid of the current repo structure with detect_fields and detect_labels to end up with a unified and cleaner layout. Therefore:

• creation of a Format class and FormatsManager to deal with them (in format.py)
• creation of a formats/ folder that contains all formats as single python files, with the format's name as file name (float.py, siren.py, ...)
• the tree structure that is used to segment the formats (FR, geo, etc.) is replaced with the tags attribute
• the labels attribute allows to merge fields and labels checks in single files, everything about let's say latitude_wgs is in latitude_wgs.py

Example:

from csv_detective.format import FormatsManager
fmtm = FormatsManager()
for format in fm.get_formats_from_tags(["fr", "geo"]):
    print(format.name)
    format.func("42")  # to test the value for each selected format

This approach will also allow to more easily perform overall additions/modifications (with in mind a new version of #155 with a parent attribute, which, combined with chunks, should speed up the analysis)

[Pierlou]

The diff is quite big, but the only files you'd like to check are:

• format.py
• formats/__init__.py
• the tests refactors if you want to check that the behaviour remains very similar after all the changes
• the previous logic that was in load_tests.py (hopefully you agree that these changes make it clearer and cleaner)

[bolinocroustibat]

Didn't review everything of course, but the current structure is definitely more readable.
A few remarks (I might add more later):

• if possible, isn't it better not to use relative imports?
• I forgot what exactly is the mechanic behind proportion (ex-PROPORTION) again. Would be great to explain it in the README (and with a little code comment as well maybe)