fix(e2e): restore synchronization in multi-VA saturation test

The multi-VA saturation test was failing with a 10-minute timeout because
the "waiting for CurrentAlloc" step had its assertions commented out when
CurrentAlloc was removed from the API (commit db784e3).

Without the assertions, the test raced against the optimization loop:
1. Test created deployments and VAs
2. "waiting for CurrentAlloc" step passed immediately (no assertions)
3. Test checked DesiredOptimizedAlloc.NumReplicas before it was populated
4. Timeout after 10 minutes waiting for a value that was never set

Fix: Replace the commented-out CurrentAlloc assertions with
DesiredOptimizedAlloc assertions, mirroring what the single-VA test
correctly does at lines 275-278. This ensures the optimization loop
has successfully processed the VAs before checking their replica counts.

Author: ev-shindin

test/e2e-saturation-based/e2e_saturation_test.go

@@ -728,7 +728,7 @@ var _ = Describe("Test workload-variant-autoscaler - Saturation Mode - Multiple
 			// 		fmt.Sprintf("VariantAutoscaling %s MetricsAvailable condition should be True", vaH100.Name))
 			// }, 4*time.Minute, 10*time.Second).Should(Succeed())
 
-			By("waiting for VariantAutoscalings CurrentAlloc to be populated with metrics data")
+			By("waiting for VariantAutoscalings DesiredOptimizedAlloc to be populated")
 			Eventually(func(g Gomega) {
 				vaA100 := &v1alpha1.VariantAutoscaling{}
 				err := crClient.Get(ctx, client.ObjectKey{
@@ -737,13 +737,11 @@ var _ = Describe("Test workload-variant-autoscaler - Saturation Mode - Multiple
 				}, vaA100)
 				g.Expect(err).NotTo(HaveOccurred(), fmt.Sprintf("Should be able to fetch VariantAutoscaling for: %s", deployNameA100))
 
-				// In saturation mode, wait for CurrentAlloc to be populated (no MetricsAvailable condition)
-				// In saturation mode, wait for CurrentAlloc to be populated (no MetricsAvailable condition)
-				// CurrentAlloc removed
-				// g.Expect(vaA100.Status.CurrentAlloc.Accelerator).NotTo(BeEmpty(),
-				// 	"CurrentAlloc should be populated with accelerator info")
-				// g.Expect(vaA100.Status.CurrentAlloc.NumReplicas).To(BeNumerically(">=", 0),
-				// 	"CurrentAlloc should have NumReplicas set")
+				// Wait for DesiredOptimizedAlloc to be populated (ensures reconciliation loop is active)
+				g.Expect(vaA100.Status.DesiredOptimizedAlloc.Accelerator).NotTo(BeEmpty(),
+					"DesiredOptimizedAlloc should be populated with accelerator info")
+				g.Expect(vaA100.Status.DesiredOptimizedAlloc.NumReplicas).To(BeNumerically(">=", 0),
+					"DesiredOptimizedAlloc should have NumReplicas set")
 			}, 5*time.Minute, 10*time.Second).Should(Succeed())
 
 			Eventually(func(g Gomega) {
@@ -754,12 +752,11 @@ var _ = Describe("Test workload-variant-autoscaler - Saturation Mode - Multiple
 				}, vaH100)
 				g.Expect(err).NotTo(HaveOccurred(), fmt.Sprintf("Should be able to fetch VariantAutoscaling for: %s", deployNameH100))
 
-				// In saturation mode, wait for CurrentAlloc to be populated (no MetricsAvailable condition)
-				// CurrentAlloc removed
-				// g.Expect(vaH100.Status.CurrentAlloc.Accelerator).NotTo(BeEmpty(),
-				// 	"CurrentAlloc should be populated with accelerator info")
-				// g.Expect(vaH100.Status.CurrentAlloc.NumReplicas).To(BeNumerically(">=", 0),
-				// 	"CurrentAlloc should have NumReplicas set")
+				// Wait for DesiredOptimizedAlloc to be populated (ensures reconciliation loop is active)
+				g.Expect(vaH100.Status.DesiredOptimizedAlloc.Accelerator).NotTo(BeEmpty(),
+					"DesiredOptimizedAlloc should be populated with accelerator info")
+				g.Expect(vaH100.Status.DesiredOptimizedAlloc.NumReplicas).To(BeNumerically(">=", 0),
+					"DesiredOptimizedAlloc should have NumReplicas set")
 			}, 10*time.Minute, 10*time.Second).Should(Succeed())
 
 			By("verifying A100 variant has expected initial replicas or scales down (before load)")