[MERGE] More robust handling of encoding issues in tests.

[Martin Pool]

Martin Pool has voted approve.
Status is now: Approved
Comment:
I think the changes to the test runner are to do with being able to 
display an error that might contain non-printable unicode messages. 
This should not normally happen but I think it did sometimes happen 
while Martin was doing other things here, and just getting a crash is 
unhelpful.

In some ways I think it would be cleaner to either use 
.encode(user_encoding, errors='replace') or just repr.  But this is 
probably reasonable.

For details, see: 
http://bundlebuggy.aaronbentley.com/project/bzr/request/%3C48733815.40704%40gmx.net%3E
Project: Bazaar