Several places in server generate sets of OperationContexts that need to be interrupted on stepdown. This is cleaner if there is a container that will interrupt all OpCtxs in it, and also tracks their lifetimes.