Extract obsolete strings

[pascalchevrel]

We don't extract obsolete localized strings in tmxmaker.py, we base our extraction on the existence of the en-US strings, therefore we don't know how many strings are obsolete in a repo which could be an interesting additional metrics on the quality of the localization. It could also allow us to compare results with the official l10n dashboards, since we don't use compare locales for comparizon, we may show up a different set of edge cases and problems.