IT: Fix and reenable flaky tests

[kaikulimu]

Summary

• Re-enabled three previously disabled flaky integration tests: test_open_queue_while_cluster_blips_quorum,
  test_open_queue_while_cluster_blips_quorum_and_kill_all_non_leader, and test_force_leader_primary_divergence.
• Fixed test_force_leader_primary_divergence by waiting for the restarted node to be ready before killing tproxies to trigger the divergence.
• Split the blips-quorum test into two variants: one where the original leader wins re-election (quorum pinned via config), and one where leadership changes (strong consistency only), each with targeted recovery logic.
• Fixed flakiness caused by cascading leader-primary divergence: when the old primary aborts after a leadership change, the new leader can lose quorum (voter count drops below threshold), triggering yet another election and a second divergence crash — leaving too few nodes for quorum.
• In mqbc::ClusterUtil, removed explit load uncommitted advs to merged state.

Future Work

• Discovered that leader-primary divergence can cause a cascading leader switch. When the old leader shuts down, it could cause the new leader to lose quorum, thus a second re-election can occur. If we are unlucky, the second leader can also abort due to leader-primary divergence. Worth discussing with @chrisbeard, @dorjesinpo, and @678098 for a proper solution.

[678098]

I think this should be done by using Configurator
For example, this is how domains configuration is updated in ITs:

    domains = {domain.name: domain for domain in cluster.configurator.domains}
    domains[
        domain_fanout
    ].definition.parameters.mode.fanout.publish_app_id_metrics = True
    cluster.deploy_domains()

[kaikulimu]

deploy_domains is a poor name. Will split the function into deploy_clusters and deploy_domains.