Add spec.groups field with typed group memberships

[PhilippMatthes]

The Cortex Placement Shim replaces OpenStack Placement for kvm and reads scheduling information directly from the Hypervisor CRD. This is an opportunity to rethink how traits and aggregates are modeled on the CRD.

Today, trait and aggregate data is split across spec and status. Controllers reconcile between the two, merging sources and pushing state to OpenStack Placement. With the shim reading the CRD directly, there is no external system to observe or merge from, making this indirection unnecessary. The existing spec fields are also flat string lists tied to OpenStack naming, unable to carry the structured data like UUIDs and metadata that the shim needs to serve a compatible Placement API.

Both traits and aggregates are forms of grouping: traits group hypervisors by capability, aggregates group them by administrative assignment. This proposal unifies them under a new spec.groups field as the single source of truth. Rather than using Kubernetes labels, which are flat key-value strings without schema validation, spec.groups models each group kind as a well-typed substruct that can carry structured data such as UUIDs and metadata.

The field is a list of typed group entries following the field-presence union pattern from core Kubernetes, as used by PodSpec.volumes. Each entry populates exactly one type-specific sub-field:

spec:
  groups:
    - trait: {name: COMPUTE_STATUS_DISABLED}
    - trait: {name: CUSTOM_TRAIT_FOO}
    - aggregate:
        name: fast-storage
        uuid: abc-123
        metadata:
          ssd: "true"

Each group type has its own struct with only the fields relevant to it. A CEL validation rule enforces that exactly one sub-field is populated per entry. The existing spec.customTraits and spec.aggregates fields remain untouched to not interfere with existing controller logic, separating this api change from the controller refactor. They can be deprecated and removed in a future release once the shim is fully integrated.

Library helper functions like HasTrait, GetTraits, HasAggregate, and GetAggregates follow the meta.IsStatusConditionTrue pattern and are reusable across the shim, operator, and scheduler.

The design is extensible: adding a new grouping concept means adding one optional pointer field to the Group struct and a corresponding type-specific struct. No changes to existing fields or consumers are required.

[PhilippMatthes]

See cobaltcore-dev/cortex#766