forEach with data.findBySpec() expands duplicate steps causing cyclic dependency error

Description

When using forEach with data.findBySpec() in a workflow, the step expansion produces duplicate steps for each version of a data entry, not just the latest version. This causes duplicate step names which triggers a "Cyclic dependency detected" error at runtime.

Steps to Reproduce

1. Create a @k8s/service model instance and run its list method a few times (producing multiple data versions per service)
2. Create a workflow with a forEach step that iterates over the data:

steps:
  - name: endpoints-${{ self.svc.attributes.name }}
    forEach:
      item: svc
      in: ${{ data.findBySpec("my-services", "service") }}
    task:
      type: model_method
      modelIdOrName: my-services
      methodName: getEndpoints
      inputs:
        serviceName: ${{ self.svc.attributes.name }}

1. Evaluate the workflow: swamp workflow evaluate my-workflow --input '...' --json
2. Observe that each service appears N times (once per data version) instead of once

For example, with 3 services and ~3 versions each, the evaluated workflow produces 9 getEndpoints steps instead of 3, with triplicate step names like:

• endpoints-frontend (×3)
• endpoints-backend-svc (×3)
• endpoints-database-svc (×3)

1. Run with --last-evaluated: swamp workflow run my-workflow --last-evaluated --json
2. Error: {"error": "Workflow execution failed: Cyclic dependency detected: "}

Expected Behavior

data.findBySpec(modelName, specName) should return only the latest version of each data entry when used in a forEach context, producing one step per unique data instance.

Actual Behavior

It returns all versions of all data entries matching the spec, causing duplicate forEach expansions and duplicate step names.

Environment

• swamp version: 20260226.174314.0-sha.0a12c9ff
• OS: macOS (Darwin 22.6.0)
• Shell: zsh

Summary

This bug affects the workflow evaluation service's forEach expansion when data.findBySpec() returns multiple versions of the same data entry. The fix would involve either deduplicating data entries by name (keeping only the latest version) during forEach expansion, or having data.findBySpec() default to returning only the latest version of each entry.