[BUG]: planner scale doesn't work if user updates deployment replicas

[sozercan]

Describe the Bug

if user changes deployments for P/D manually, planner reports on deployment replicas correctly

│ trtllm-disagg-planner-5777d79b79-pjblx 2025-11-17T22:18:42.259679Z  INFO planner_core.make_adjustments: Number of prefill workers: 2, number of decode workers: 2      │

but planner doesn't work to scale down, since DGD is the source of truth

2025-11-14T19:47:06.040464Z  INFO kubernetes_connector.set_component_replicas: prefill component TRTLLMPrefillWorker already at desired replica count 1, skipping
2025-11-14T19:47:06.040538Z  INFO kubernetes_connector.set_component_replicas: decode component TRTLLMDecodeWorker already at desired replica count 1, skipping

Steps to Reproduce

• update replica count for P/D in k8s deployment (not dgd)
• dont send traffic to dynamo
• watch planner not being able to scale down

if user updates dgd, it will scale down properly

Expected Behavior

planner scales down or tells user that they are holding it wrong

Actual Behavior

planner cannot scale down

Environment

dynamo 0.6.1

Additional Context

cc @tedzhouhk

Screenshots

No response