Yet another version of the enter/exit scripts to get DHCP to work both
regular- and elabinelab-wise.  For regular experiments, nothing has changed.
New strategy for inner elab nodes: if we get a reply with dns server 1.1.1.1
(our current indication that the responder is real boss), then we know this
interface is the real control net.  So we hardwire the speed/duplex on
*every other* interface at this time.  Also had to resurrect a hack to
issue a few tmcc calls before running the sethostname script.  Apparently,
the first few DNS queries after a speed/duplex change may get lost.  This
causes tmcd calls to fail occasionally.  So make em fail early before we
try to do anything important.

Yuk, yuk, yuk.