Rename message_id to time_id. Use _id for sequential message index given by producer and time_id for message based on arrival time.

This includes read-only buffers.

The code split between bytes and numpy array is removed and all objects
are now handled equally.

Furthermore, it seems to be more correct to increase the ref count for
None objects as well, while a pointer to that object is used.

Enable test of receiver with kafka. Add sleep of 3 sec after start of kafka listener to ensure that is it ready