Collapse compute node and ood ops provisioners into a single
null_resource to avoid a race condition when the ansible tasks
run in parallel.  This lead to inconsistent deploys where
the slurm configuration was not working after deploy.

Putting the provisioners in a single resource ensures they run in
series.  This comes at the cost of a loss of deploy flexibility.
Now ood and compute nodes ops steps will always run on changes rather
than selectively.  Since this feature wasn't fully functional, it's
loss is not significant.