Add CLDR data source for subdivion translations.

[rposborne]

This massive improves the number of translations we have available and lays the groundwork of other importers.

Thanks to @coderifous and @mig for the help on this.

This fixes #437

[jawnsy]

I perused these changes quickly and they look good to me. I think it's a better approach than what I proposed with #438 because it handles updates better (regrettably I didn't understand the way updates are currently handled and couldn't do as good a job as you did there).

Unfortunately I can't test the changes as I left Red Hat and no longer have access to the source code of the project I was working on.

One thing I noticed is that many fields where the data is presumably unknown are recorded as empty fields, and I wonder if it would be better to exclude them from the file entirely? For example, in AM.yaml:

  geo:
    latitude: 
    longitude: 
    min_latitude: 
    min_longitude: 
    max_latitude: 
    max_longitude: 
  name: Gegark'unik'

It seems the whole geo section can be excluded? Over all of the input files this might save some disk space, though probably not a huge amount, and it doesn't matter much with compression.

[rposborne]

@jawnsy it looks like geo is filled out for most sub divisions. You make an interesting case though to remove fields that have nil values. Or pass the cache through a compression layer, which currently it is not. I would be curious to see what gives us a better performance gain.

I think for now this is focused on data integrity and data import over the consumption of the data. So I'll break those out into separate issues.

[jawnsy]

@rposborne Sounds like a reasonable approach to me! Thanks for your work!

[rposborne]

@jawnsy 2.1.3 released which includes this change