Check categories are mixed up after subsequent runs

[julen]

The check categories are assigned to the class upon respective method execution via the decorators. This is problematic if some code wants to read the categories for a specific checker after another checker has run, because the class attribute is shared across instances.

This breaks code/tests in Pootle, e.g. get_qualitychecks.

The weird behavior can be demonstrated with this test, which is expected to fail:

def test_category():
    """Tests checker categories aren't mixed up."""
    from translate.storage import base
    unit = base.TranslationUnit(u'foo')
    unit.target = u'bar'

    standard_checker = checks.StandardChecker()
    assert standard_checker.categories == {}
    standard_checker.run_filters(unit)
    assert standard_checker.categories != {}
    assert 'validxml' not in standard_checker.categories.keys()
    standard_categories_count = len(standard_checker.categories.values())

    libo_checker = checks.LibreOfficeChecker()
    assert libo_checker.categories == {}
    libo_checker.run_filters(unit)
    assert libo_checker.categories != {}
    assert 'validxml' in libo_checker.categories.keys()

    standard_checker = checks.StandardChecker()
    assert standard_checker.categories == {}
    standard_checker.run_filters(unit)
    assert standard_checker.categories != {}
    assert len(standard_checker.categories.values()) == standard_categories_count
    assert 'validxml' not in standard_checker.categories.keys()

Leaving aside possible improvements to how tests are categorized and executed, a less surprising behavior would be to store the categories on instances. I will PR a patch in this direction.