Don't normalize translations stored in the translation memory

The normalization mode previously used by gtranslator when adding to
the translation memory decomposes accented characters into separate
combining characters. This can cause problems where some fonts render
the decomposed characters differently. In particular Cantarell has
problems rendering circumflex accents.

As far as I can see the only reason you might want to normalize the
strings would be when comparing them. However the only place I can see
where the translations are compared is when the translation is
removed. In this case the string compared against is directly taken
from the same database so it should always be exactly the same.

This patch therefore just removes the normalization altogether. If we
did want to compare strings for any other reason I think we should
either store both versions of the translation or try to do the
normalization immediately before the comparison. Modifying the strings
coming from other .po files at all seems like a bad idea.

https://bugzilla.gnome.org/show_bug.cgi?id=660813