BlueGreen doesn't scale down previously active replica after a rollback

[samuelbaena]

Describe the bug

After performing a rollback in Argo Rollouts, I would expect the previously active replica to downscale to 0 once the autoPromotionSeconds, scaleDownDelaySeconds as well as any analysis time happens, this is not true as both the rollbacked version and the until then active/stable one remains

To Reproduce

blueGreen:
      abortScaleDownDelaySeconds: 30
      activeMetadata:
        labels:
          role: active
      activeService: demo-test
      autoPromotionEnabled: true
      autoPromotionSeconds: 30
      previewMetadata:
        labels:
          role: preview
      previewReplicaCount: 1
      previewService: demo-preview
      scaleDownDelaySeconds: 30

1. Deploy normally
2. Rollback to previous version

Expected behavior

The previously Active replica is downscaled.

Screenshots

Version

v1.6.0

---

Message from the maintainers:

Impacted by this bug? Give it a 👍. We prioritize the issues with the most 👍.

[samuelbaena]

The team figured out that we were affected by this issue
Bumping sorted it

[samuelbaena]

Worked once, broke once. Difficult to identify a pattern on which times it happens, but it has happened again with 1.6.4