What type of PR is this?

/kind feature
/kind api

What this PR does / why we need it:

Implementation of Multiple Topology GREP to enhance TAS support (#413 )

Which issue(s) this PR fixes:

Fixes #369

Special notes for your reviewer:

Does this PR introduce a API change?

ClusterTopology API extended with multi-topology support:

ClusterTopologySpec:

• spec.levels[].domain — removed fixed enum constraint (was region|zone|datacenter|block|rack|host|numa); domains are now free-form strings validated by pattern ^[a-z][a-z0-9-]*$ with max 63 characters. This enables heterogeneous GPU cluster topologies where hardware segments define their own domain names.
• spec.schedulerReferences (new, optional) — list of {schedulerName, reference} entries. Controls per-backend topology resource lifecycle: absent = operator auto-manages the backend topology resource; present = resource is externally managed and the operator performs drift detection only.

ClusterTopologyStatus (new):

• status.observedGeneration — generation last reconciled by the controller.
• status.conditions — standard condition list; includes SchedulerTopologyDrift condition (True/Drift when any backend is out of sync, False/InSync when all match).
• status.schedulerTopologyStatuses — per-backend sync state reporting {schedulerName, reference, inSync, message, schedulerBackendTopologyObservedGeneration}.

PodCliqueSet API:

• spec.template.topologyConstraint type changed from TopologyConstraint to PodCliqueSetTopologyConstraint, adding topologyName field — the name of the ClusterTopology resource to use. Required when packDomain is specified. Immutable after creation.
• spec.template.topologyConstraint.packDomain enum constraint removed; must now reference a domain defined in the named ClusterTopology's levels (validated by webhook).

Removed:

• ClusterTopology.spec.levels immutability constraint and MaxItems=7 cap.
• DefaultClusterTopologyName ("grove-topology") constant — the operator no longer manages a single implicit topology; all topologies are now admin-created ClusterTopology resources.
• TopologyDomain.IsTopologyDomainNarrower, SupportedTopologyDomains, SortTopologyLevels helper functions — ordering was predicated on the fixed enum which no longer exists.

Conditions added (constants):

• SchedulerTopologyDrift / InSync / Drift — on ClusterTopology
• TopologyNameMissing / TopologyAwareSchedulingDisabled — on PodCliqueSet

> [!WARNING]
> **Upgrade Notice — Topology-Aware Scheduling**
>
> Affects users who have deployed `PodCliqueSet` resources with a `topologyConstraint` using the previous single-topology design.

### What changed

The operator now supports multiple named `ClusterTopology` resources. Each `PodCliqueSet.spec.template.topologyConstraint` now requires a `topologyName` field that explicitly names the `ClusterTopology` to use. Without it, topology placement constraints are not enforced.

### Action required

Add `topologyName: grove-topology` to each existing `PodCliqueSet` that has a `topologyConstraint`:

spec:
  template:
    topologyConstraint:
      topologyName: grove-topology   # add this
      packDomain: rack

The existing `grove-topology` ClusterTopology resource created by the previous operator version remains in the cluster and is valid — no changes to it are needed.

### What happens if you don't act immediately

The operator degrades gracefully but **topology placement constraints will not be enforced** until `topologyName` is set. This is visible via the `TopologyLevelsUnavailable` condition on the PCS:

| Condition status | Reason | Meaning |
|---|---|---|
| `Unknown` | `TopologyNameMissing` | PCS has `topologyConstraint` but no `topologyName`. Constraints are not enforced. Add `topologyName` to restore them. |
| `True` | `ClusterTopologyNotFound` | The named `ClusterTopology` does not exist. Create it or correct the name. |
| `False` | `AllClusterTopologyLevelsAvailable` | All good — constraints are being enforced. |

Check for affected PodCliqueSets after upgrade:

kubectl get podcliquesets -A -o json | jq -r '
  .items[] | select(.status.conditions != null) |
  .status.conditions[] | select(.type == "TopologyLevelsUnavailable" and .status != "False") |
  "\(.type): \(.reason) — \(.message)"'

Additional documentation e.g., enhancement proposals, usage docs, etc.:

TBA