We still had some lingering type problems distinguishing between unicode and

bytes in parsing git diffs. This meant that diffs that including utf-8

characters would cause UnicodeDecodeErrors when type coercion tried to

interpret utf-8 encoded text as ascii. One example of this is the diff attached

to /r/3749/.
This change makes sure that during the diff parse, we're always treating the

text as bytes. While this is theoretically bad for non-utf8 encodings, it's no

worse than we've always had, and makes it so we can actually parse these diffs.

Loaded /r/3749/. Saw the page instead of a traceback. Verified that the raw

diff returned the right utf-8 encoded patch.