Dynamically adjust slack quota

[tardieu]

This PR makes it possible to designate a cluster queue as slack, for example:

    appwrapper:
      enableKueueIntegrations: true
      defaultQueueName: default-queue
      slackQueueName: slack-cluster-queue
      autopilot:
        injectAntiAffinities: true
        migrateImpactedWorkloads: true
        resourceUnhealthyConfig:
          nvidia.com/gpu:
            autopilot.ibm.com/gpuhealth: ERR

The AppWrapper controller will dynamically inject and adjust lending limits on the resources in this cluster queue. These limits are computed by subtracting unhealthy resource counts (as reported via autopilot labels) to the nominal quotas for these resources.

For instance, if an eight-gpu node reports an error (via label autopilot.ibm.com/gpuhealth: ERR), the following lending limit is injected in the slack cluster queue:

     resources:
      - name: cpu
        nominalQuota: "8"
      - name: memory
        nominalQuota: 128Gi
      - lendingLimit: "16"
        name: nvidia.com/gpu
        nominalQuota: "24"