Feature Request: Support DV-level scheduling constraints (nodeSelector/affinity/tolerations) for importer pods

[jimyag]

Is your feature request related to a problem? Please describe

Currently, scheduling constraints for CDI workload pods (importer, cloner, uploader) can only be configured globally via CDI.spec.workload. There is no way to specify different scheduling requirements for individual DataVolumes.

When using local storage provisioners, we need to control which node the importer pod runs on at the DV level.

Describe the solution you'd like

Add a workloadPlacement field to DataVolumeSpec, limited to scheduling-related fields only:

  apiVersion: cdi.kubevirt.io/v1beta1
  kind: DataVolume
  spec:
    source: ...
    pvc: ...
    # New field - limited to scheduling only
    workloadPlacement:
      nodeSelector: {...}
      affinity: {...}
      tolerations: [...]

This approach:

1. Only exposes scheduling fields, no env, volumes, or other risky fields
2. Reuses the same NodePlacement struct pattern as CDI.spec.workload
3. Optional field; existing DVs continue to use global config

Precedence: DV-level workloadPlacement overrides CDI.spec.workload for that specific DataVolume. If DV-level is not specified, fall back to global config.

Describe alternatives you've considered

Solution	Limitation
CDI.spec.workload	Cluster-wide only; cannot vary per DataVolume
WaitForFirstConsumer	Waits for consumer pod, but virt-launcher won't start until DV is ready (chicken-and-egg)
PV node affinity	Still in alpha; not considered production-ready
Third-party solutions (e.g., mutating webhooks)	Adds external dependencies and complexity

We believe it would be better to leverage the native capabilities of DataVolume and importer pod directly, rather than relying on external workarounds.

Additional context

This issue is split from #3848 to discuss DV-level scheduling constraints separately from scratch space storage class configuration.