We use pybabel to extract strings from third-party apps, put them on
Transifex, then merge them into django.po.

[LMS-1681]

Now normal i18n:generate doesn't care about missing files, and
i18n:generate_strict does.  The robot task calls the strict version.

There may well be a better way to make the two tasks, but my Ruby+Rake
skillz are weak, and it's all going away soon.

Add section headings to test console output

i18n tests generate XUnit reports

Move i18n and docs test commands out of tests.rake

Removed unused make_dummy.py
Made dummy.py executable

LMS-1967

We were writing the main file twice.  No need.

When this was conditional, we'd have plural info in some files, not in
others. Then we'd merge them, and msgcat would add wacky markers to
indicate the difference, and the metadata wouldn't parse.

Use the configuration more, and we don't need to delete files we're
about to create.

Messages can be segmented into different .po files based on the location
of their occurrences.  This is so Studio strings can go into different
.po files so we can direct Transifex workers to the LMS strings we
really need done.

LMS-1963

Now any of the plurals being missing will count as an Empty translation,
and for looking at tags, the two English strings and all the translated
strings are mooshed together.

The tests in test_validate.py now examine the messages to see that they
have the right structure.  The English and the translation should have
the same tags, the translation can't be blank, and it can't have any
Astral plane characters.