This annoying bug affects unit tests. Since we're just printing out random byte
strings from random HTTP endpoints, there is no guarantee that we're getting
text. And we often don't, when we do file I/O with WebHDFS. This causes us to
log non-UTF8 safe stuff, which makes nosetest put non-UTF8 data into the XML
report file, which means that Jenkins can't parse it.

(cherry picked from commit 5d13ff8c695792fb9bc090cd61c8d8987afd9f54)
(cherry picked from commit e2e788bbab82a004fc9e35e17fe5ba2cd6645b0c)