fix: wait for block node pod termination before checking new pod readiness

waitForReadyStatus returned immediately seeing the still-Ready old pod after
deletion, completing deployment while the new pod was still starting.  This
caused the consensus node to exhaust its gRPC publisher reconnect attempts
against a non-existent endpoint and never reconnect in the 5-minute window
before PUBLISHER_UNAVAILABILITY_TIMEOUT fired.

Add waitForPodsToTerminate between delete() and waitForReadyStatus() so that
the new pod is the one polled for Ready status, giving the CN time to
reconnect before the block node's publisher unavailability timer expires.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: Jeffrey Tang <jeffrey@swirldslabs.com>

Author: JeffreyDallas

src/commands/one-shot/orchestrator/deploy/default-one-shot-deploy-orchestrator.ts

@@ -804,6 +804,18 @@ export class DefaultOneShotDeployOrchestrator implements OneShotDeployOrchestrat
 
             // Restart the pod so BlockNodeApp.loadApplicationState() picks up the file on startup.
             await k8.pods().delete(blockNodePod.podReference);
+            // Wait for old pod to fully terminate before waiting for the new pod.
+            // Without this, waitForReadyStatus returns immediately seeing the still-Ready old pod,
+            // completing deployment before the new pod exists and causing the consensus node to
+            // exhaust its gRPC publisher reconnect attempts against a non-existent endpoint.
+            await k8
+              .pods()
+              .waitForPodsToTerminate(
+                namespace,
+                blockNodeLabels,
+                constants.BLOCK_NODE_PODS_RUNNING_MAX_ATTEMPTS,
+                constants.BLOCK_NODE_PODS_RUNNING_DELAY,
+              );
             await k8
               .pods()
               .waitForReadyStatus(