This is based on 2.8.2.2 from ECMA OOXML Part 4 and is a bit of a guess.
I do not see why JIS should be the charset for everything that is unicode,
and MSO2013 fails to read embedded fonts with w:charset set to this value.
It seems to handle them when w:charset is set to this 'default character
set', whatever that actually is. Without embedded fonts involved it moreover
looks like w:charset does not actually matter.

Change-Id: I7c083a0e39076313cd2687b1712420c92f691f5b