> While reading through [1] I realized that the word boundary detection
> also uses the wrong collation. Patch 0002 fixes that.
Good catch, I considered checking that when I reviewed the patch, but
I thought that it would be an issue with CJK languages, and I was
completely wrong about that.
Updated patch looks good, I only noticed two minor things:
* the new test should have a newline at the end of the file
* and probably a conditional skip based on locale availability, the
citext_utf8.sql test case does something similar
