This PR moves the underlay teardown logic from xde_detach into a new
ioctl (and opteadm command) to unset the XDE underlay. Detach will now
fail so long as the underlay is set -- this successfully prevents
`modunload` from knocking us down after this state has been set.

The semantics are more or less what you'd expect:
* Clearing the underlay will fail if there exists a port.
* Clearing the underlay will fail if there is no underlay.
* Driver detach now requires that you `opteadm clear-xde-underlay`.

Following discussion, we need to figure out a safer quiesce on rx as
there is a theoretical bad ordering between `xde_rx` and underlay
removal.

After some research into MAC promiscuous callback behaviour, I've
verified that we don't need to do our own quiescing since the cleanup
operations will await any active promisc list walkers terminating. I've
explained this logic for the benefit of others and/or future Kyle.

Minor papercut I just ran into, this should keep the command
zero-friction.