While treating output as text would not really be impossible, treating it
as bytes seems a better choice because:
 - we don't have to make assumptions about what output encoding the test
   program is using for output
 - `tee` can just read stream output bytes by bytes without having to worry
   about multi-bytes characters
 - testnode protocol uses xmlrpc.client.Binary, which uses bytes.

Because using bufsize=1 implies reading subprocess output as text, we use
bufsize=0 instead in the subprocess.Popen call, to prevent buffering.

To make manipulation of strings and bytes easier, we add a dependency on
pygolang, so that we can use its strings utility functions.

Also add a few tests to verify general functionality.