error instead of doing an assertion failure.

move some duplicated logic into a common function

    nodes already in another reservation.

  from the database

we're able to handle this now with the new logic that tests scheduling with 
preempted jobs removed from our simulated scheduling matrix