The diff viewer code, which is used for text-based attachment diffs,

expects data to be converted to Unicode for display, and to do this, we

run the text through convert_unicode(). However, this only happens if

there's an encoding list set up, which there isn't by default. Diffs

backed by repositories do this because we always have a default encoding

list, but there was no such logic here.
Now, the diff viewer provides a default for the encoding list, and it

also unconditionally runs convert_unicode(). This allows text-based

attachments with Unicode characters to properly render.

Some file attachments I tried that were failing now properly render.