There are a few issues that have been cropping up with the diff parsing (and

other interfacing with output from 'svn' commands) since we moved over to

unicode_literals. This change fixes three of them.
First of all, the results from 'svn info' can include non-UTF8 characters

depending on the locales that are available and the configuration of the

system. Luckily, all of the information we need from this can be parsed out in

binary mode.
Second, when doing diffs of files that contain non-UTF8 characters, we were

attempting to process the diff as a bunch of unicode objects, rather than as

bytes. This was happening because we'd format some of the lines into unicode,

and then join everything together. I've fixed things so that all the processing

happens as bytes, so that we don't need to care about the encoding, and added a

new unit test for this behavior.
Finally, Griffin Myers suggested a third test case, with a non-utf8 filename.

This has also been fixed, along with a new unit test for that case.
Testing done:

Ran unit tests.

Ran unit tests.