#604 Dynamic resource attribute refresh for model instances

Problem

When recovering infrastructure through swamp model instances (e.g., Kubernetes nodes or cloud VMs), the model's resource attributes are stale — they reflect the state at the time of the last sync or lookup, not the current live state. This makes it difficult to use swamp workflows for recovery operations where resource attributes change between runs.

Example scenario

A cluster has 5 nodes. After a failure, 2 nodes are destroyed and recreated by an IaC tool. The recreated nodes have new IP addresses. A swamp workflow needs to SSH to these nodes to perform recovery steps (wipe data directories, restart services). The workflow references a node model instance, but the model still has the old IPs from before the recreation.

The operator has to:

1. Manually run sync on the model to refresh attributes
2. Wait for the sync to complete
3. Then run the recovery workflow

This breaks the single-workflow recovery pattern. In practice, operators fall back to direct CLI (ssh, kubectl) because the model data is stale and there is no way to force a refresh mid-workflow.

Proposed solution

One of:

• A refresh step type in workflows that forces a model sync before continuing
• Lazy evaluation of model attributes at step execution time (re-read from the provider on access rather than using cached data)
• A depends_on_sync flag on workflow steps that triggers a sync before the step runs

Workaround today

Run swamp model method run <name> sync manually before the recovery workflow. This works but requires operator intervention and breaks automation.