This is very similar to what Emulab does on the swapin path for
normal experiments; wait and watch the nodes to see which ones
fail or otherwise timeout. Up till now, we did not do this on the
PG path, and so failed nodes were never signaled, and the slice
was left in a changing state forever. This also allows us to capture
the node bootlogs and convert them to logfiles that we can associate
with the slice on the showslice web page.

Details: start/restart forks a child (WrapperFork()) and allows
the parent to return to the client. The slice is unlocked so that
the client can call SliverStatus(), etc. But the client cannot
do anything that actually changes the sliver (update, stop, etc)
until the monitor finishes (or times out on its own). The lone
exception is Deleteslice(), which will asynchronously kill the
monitor and then terminate the slice. Ditto the command line
script "cleanupslice".

We will probably need to add another way to allow the client to
terminate the monitor early, but have not decided where yet.