Until now, we've been assuming that all source files are either UTF-8
or ISO-8859-15.  This breaks things for people who use other, incompatible
encodings like Shift-JIS.

This change adds an optional encoding field to the repository model so
people can set it there.  This is probably the best we can do, since I
don't know of any VCS that actually keeps track of the encoding of files
stored inside it.