[RI-20] Populate Kubernetes topology labels from Oxide failure domains

[sudomateo]

The CCM’s InstanceMetadata implementation returns provider ID, instance type, and node addresses but omits zone and region metadata. Kubernetes consequently receives no topology.kubernetes.io/zone label and cannot use scheduler topology constraints to spread workloads across Oxide failure domains.

Populate a stable, distinct topology value for each Oxide failure domain through InstanceMetadata.Zone. This may depend on the Oxide API exposing a tenant-visible instance-to-failure-domain identifier; coordinate that dependency rather than assigning every node the generic value sled.

[sudomateo]

Definitely something we want to do, though we'll need to decide what a region and zone actually mean in the limit. I think we can take a half-measure here and populate labels for the rack ID and sled ID.