The patches changes:

- change PyString's repr() to return "b'...'"
- change PyBytes's repr() to return "buffer(b'...')"
- change parser so that b"..." returns PyString, not PyBytes
- rename bytes -> buffer, str8 -> bytes

The patch breaks some unit test because some of the infrastructure like
the new buffer() isn't in place yet. I'm not happy with bytes_repr() but
the for loop with *p++ was the easiest way to implement it. Every other
implementation I could think of was either too complicated or wouldn't
work (like memcpy).