kubernetes_nodes: expose maxUnavailable so a pool without spare capacity can drain first #79

cozystack/cozystack#5153 adds maxUnavailable to the KubernetesNodes spec (packages/apps/kubernetes-nodes), and to each worker pool of a ComputePlane, which forwards it to the pool chart. Unset, it is 0: a rollout (instance type, image or version change) brings up each replacement worker before it drains an old one. A pool with no room for one extra VM, short of quota or capacity, then stalls with the new Machine Pending, and the documented way out is to set maxUnavailable: 1, which drains first.

cozystack_kubernetes_nodes cannot set it. internal/provider/kubernetes_nodes_schema.go has no attribute for it, and since an update replaces the whole spec, a value set outside Terraform is dropped on the next apply. A Terraform-managed pool that needs to drain first therefore has no way to express it.

What is needed:

• a max_unavailable attribute on the cozystack_kubernetes_nodes resource and data source, Optional, Int64, between 0 and 2147483647 (the bounds the API schema enforces);
• no provider-side default: left unset it must not be sent, so the chart keeps choosing 0, or 1 on a pool with gpus;
• the field in the model and the expand/flatten pair.

The provider has no ComputePlane resource, so nothing is needed there yet; whoever adds one should carry the per-pool maxUnavailable too.