[CASCL-1386] Add e2e tests for autoscaling cluster evict-legacy-nodes

[L3n41c]

What does this PR do?

Adds TestEvictLegacyNodes to the autoscaling e2e suite (test/e2e/tests/autoscaling_suite/), giving the kubectl datadog autoscaling cluster evict-legacy-nodes subcommand its first end-to-end coverage. The subtests run sequentially on the shared EKS cluster, in a deliberately ordered flow:

1. Refuses when Karpenter not installed — asserts the precondition error (run with --dry-run so a violated precondition can never destructively drain the cluster).
2. Validation errors — table-driven checks for mutually-exclusive --all/--target, missing target, non-positive --eviction-timeout, and the rejected fargate target (all fail in validate() before any cluster call).
3. Install — default install (Karpenter + Datadog-managed NodePools).
4. Dry-run does not mutate — --all --dry-run produces a plan, cordons nothing, and leaves the legacy managed node groups intact.
5. Drain all legacy node groups — real --all eviction; verifies the EKS managed node groups are drained to zero and every workload pod reschedules onto a Datadog-managed Karpenter node.
6. Evict is idempotent — a second --all run is a no-op (Nothing to evict).
7. Uninstall — restores the clean post-suite state.

Motivation

evict-legacy-nodes (CASCL-1386) was merged across ~a dozen PRs with thorough unit coverage but no e2e coverage, unlike its siblings install / uninstall / update. This closes that gap and validates the orchestrator end-to-end (preconditions, dry-run safety, the real cordon→drain→scale-to-zero→reschedule path, and idempotency) against a live cluster.

Additional Notes

• The test reuses the existing suite helpers (runKubectlDatadog, testInstall, testUninstall, waitForAllPodsRunning, waitForPendingPods) and the shared anti-affinity workload; a new runEvict helper centralizes the CLI prefix. It reuses the production constant aws.LabelEKSNodegroup so the test stays coupled to the label the drain code actually keys off.
• The single-method-with-subtests shape is intentional: the ordering is load-bearing (dry-run must run while the node groups still have nodes; the destructive drain must run after it).
• Known limitation: the drain subtest scales the EKS managed node groups to zero and does not restore them. This is safe because the e2e framework provisions an ephemeral cluster per run; a reused cluster would start the suite with no managed-node-group capacity. Documented inline.
• No CI/build-tag changes needed — the e2e_autoscaling job's changes rule already covers test/e2e/tests/autoscaling_suite/**/*, and make e2e-autoscaling-tests builds bin/kubectl-datadog first.

Jira: CASCL-1386

Minimum Agent Versions

Are there minimum versions of the Datadog Agent and/or Cluster Agent required?

• Agent: n/a
• Cluster Agent: n/a

Describe your test plan

• go vet and gofmt pass on the new file.
• The suite cannot run locally (requires EKS/AWS); it is validated by the e2e_autoscaling GitLab CI job, which builds the plugin and runs make e2e-autoscaling-tests. The new subtests add roughly one extra install + drain + uninstall cycle to the suite runtime.

Checklist

• PR has at least one valid label: bug, enhancement, refactoring, documentation, tooling, and/or dependencies
• PR has a milestone or the qa/skip-qa label
• All commits are signed (see: signing commits)

🤖 Generated with Claude Code

[datadog-datadog-prod-us1-2]

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 46.29% (+0.00%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: e92e2b2 | Docs | Datadog PR Page | Give us feedback!}