Map dummy language codes in sites

Map dummy language codes in sites

The script that populates the sites table assumes that the
subdomain is the content language code. This is not true for
all wikis. This patch introduces a mapping to fix this issue
on the fly, based on $wgDummyLanguageCodes.

This is driven by the need to avoid "bad" language codes when
creating wikidata items from the client, when "linking" two
pages on different wikis. When we do this, we use the language
code from the sites table for the labels of the new item.
We would like to forbid "dummy" languages in labels and
descriptions, for consistency.

Change-Id: I6452761e14d9902bb069e32d0f499bc39e680453