When issue64 has been solved and metadata is treated as Unicode, we should also
do matching well. So doing something like "darcs changes --match 'author 乮丽
亦+'" should work as one expects.

Right now, when reading command line arguments, the Haskell runtime treats them
as latin1. That breaks matching.

Because we decided that darcs will not store its patches in a normalized
form, the matching will have to do be normalization-aware.

Perhaps we have to add a compile-time option to disable this
normalization awareness because on some platforms it is apparently hard
to use Unicode libraries from Haskell.