[ofw] running ibnetdisc with the debugger attached

[Hefty, Sean]

> I must say that once we have been running with the debugger we have found
> that the function fails unless we zero offset field.

I've committed a change where the entire overlap structured is zeroed on creation.  The overlapped documentation states that "the system will not modify these values" in reference to the offsets, so I believe this should be sufficient and avoid the overhead of setting the values on every Send/Receive call.