Always use tokenize.open instead of chardet to detect the encoding.
We meant chardet for Python 2, although theoretically it might
be needed in corner cases on Python 3. Let's try without it.
Use shlex.quote instead of our simpler quoting in format_command.
The main ones: