[MERGE] UTF-8 encoding in binary diffs

[Martin Pool]

Martin Pool has voted -1.
Status is now: Vetoed
Comment:
I think changing to %s rather than %r is definitely right.

And I think encoding just the filename, rather than the whole stream, is 
also probably right, as it will give better results if we don't know the 
encoding of the contents of the file.

UPDATE: Not mergeable in its current form because it breaks when the 
test case can't create unicode filenames - jml, could you please fix 
that as described on the list?
My only query is whether we should be hardcoding utf-8 here.  Shouldn't 
we be putting the filenames into the user's encoding?

For details, see: 
http://bundlebuggy.aaronbentley.com/request/%3Cd06a5cd30707090048l588e49c3wdfb8aaf3379f3b5e%40mail.gmail.com%3E