[Bug]: Orphaned driver pod migration sets upgrade-required on NVIDIADriver pools with autoUpgrade disabled #3063

Describe the bug
labelNodesWithOrphanedDriverPods (controllers/nodelabeling_controller.go:581) sets the upgrade state label to upgrade-required (line 625) on any node owned by a live NVIDIADriver that runs an ownerless driver pod. It does not check that NVIDIADriver's upgradePolicy.autoUpgrade.

For an NVIDIADriver with autoUpgrade: false, reconcileNVIDIADriverUpgrades removes the same label (controllers/upgrade_controller.go:262). Each later node-labeling reconcile adds it back, so the label flaps and the orphaned pod is never replaced.

To Reproduce
Found by code review; not yet reproduced. Expected: migrate from a ClusterPolicy-managed driver to an NVIDIADriver with upgradePolicy.autoUpgrade: false, leaving the old driver pods orphaned; watch the node's upgrade state label across reconciles.

Expected behavior
Skip nodes whose NVIDIADriver has auto-upgrade disabled, and report (event/condition) that the orphaned pod needs manual replacement.

Environment (please provide the following information):

• GPU Operator Version: main @ f806d6b (code inspection)