Nice, this method allows to reduce the redundancy to 

      job_clusters:
        - job_cluster_key: small_cluster
          new_cluster: ${var.small_cluster_id}  

which has to be repeated within each job definition. But this redundancy is still a lot! How can we define a job_cluster exactly once and refer to it by name / id? In other words, what is the job-cluster equivalent to the top-level resource cluster?