Before this change the s3 backend used Dot in its encoding. This meant
it was impossible to read and write a file called "." as we were
translating it to its unicode equivalent.
This means that the integration tests weren't testing the ability to
save and load files called "." and ".." at all.
This change removes the encoding which means that files called "."
(which need to be named with the unicode equivalent "．" on the
command line) can be used.
c1bf3f3999