Don't use substring matching on locales.

We can't assume that all languages will be 2 chars in length.
Language matching will need to parse the language subtag from
the locale, else we might end up matching a languages like
"fi" (finnish) to "fil" (filipino).

bug: 10090157

(cherry picked from commit f72d7c26)

Change-Id: I7139ddc99dd7d71f82ecb2c1e77af077b659a5a5