When we read files from disk we always want to read them as bytes and
then convert them to text ourselves. This gives us greater control over
what encodings are used, what to do in case of errors decoding the bytes
to text, and better resilience against problems on both Python 2 and
Python 3.
If we left it up to Python to do this, on Python2, this could mean
that config values end up as bytes (leading to tracebacks elsewhere in
the code). In Python3, it could traceback if the user's locale did not
match with the encoding of the ini file or config files could be decoded
as the user's locale encoding instead of as utf-8.