fix(e2e): Clear stale bootstrap IP annotation in bare metal API upgrade flow (#10652)

The CLI-generated config file is written before the bootstrap-to-management
cluster pivot, so it retains the tinkerbell-bootstrap-ip annotation. When
the test re-applies this config via kubectl apply, the annotation gets
re-added to the Cluster CR, causing the controller to regenerate machine
templates with the bootstrap IP instead of the tinkerbellIP. This creates
a template diff that triggers unintended rollouts, blocking reconciliation.

Co-authored-by: Rahul Ganesh <rahulbabu95@gmail.com>

Author: eks-distro-pr-bot

test/e2e/upgrade.go

@@ -134,6 +134,12 @@ func runUpgradeFlowForBareMetalWithAPI(test *framework.ClusterE2ETest, fillers .
 	test.GenerateHardwareConfig()
 	test.CreateCluster(framework.WithControlPlaneWaitTimeout("20m"))
 	test.LoadClusterConfigGeneratedByCLI()
+	// The CLI-generated config file is written before the bootstrap cluster pivot,
+	// so it still contains the tinkerbell-bootstrap-ip annotation. This annotation
+	// causes the controller to use the bootstrap IP when regenerating machine templates,
+	// creating a diff with the existing templates (which were updated during pivot to
+	// use the tinkerbellIP). Clear it so the kubectl apply doesn't re-add it.
+	test.ClusterConfig.Cluster.ClearTinkerbellIPAnnotation()
 	test.UpdateClusterConfig(fillers...)
 	test.ApplyClusterManifest()
 	test.ValidateClusterState()