fix: migrate NodeChurnSpec to async TestKit APIs to fix shutdown-timeout flakiness

[Aaronontheweb]

Problem

NodeChurnSpec intermittently fails on CI with:

System.TimeoutException : Failed to stop [NodeChurnSpec] within [00:00:05]. ActorSystem is being forcefully shut down.
   at Akka.TestKit.TestKitBase.Shutdown(...)

(Most recently surfaced on the Microsoft.NET.Test.Sdk 17.9.0 → 18.7.0 bump in #8282, which is purely an environmental/test-runner timing change — not a product regression.)

Root cause

The spec creates two transient ActorSystems per round and tears them down with the synchronous Shutdown(node, verifySystemShutdown:true) helper, which does system.Terminate().Wait(5s). That blocks a thread-pool thread for up to its 5s budget while the coordinated-shutdown pipeline (run-coordinated-shutdown-when-down = on → cluster leave/down + remoting teardown + scheduler stop) needs that same thread pool to make progress. On slower/contended CI agents this sync-over-async self-starvation pushes a normally-fast shutdown past the hardcoded 5s default and the verify path throws.

Fix

Migrate the spec to the async, task-returning TestKit APIs and stop blocking on shutdown:

• [MultiNodeFact] void → async Task (already the dominant pattern; sibling specs LeaderElectionSpec / ClusterAccrualFailureDetectorSpec in this project are async)
• Within/AwaitMembersUp/EnterBarrier/AwaitAssert/ExpectNoMsg → WithinAsync/AwaitMembersUpAsync/EnterBarrierAsync/AwaitAssertAsync/ExpectNoMsgAsync
• Replace the blocking per-system Shutdown(...) loop with concurrent async termination:

  await Task.WhenAll(systems.Select(s => s.Terminate())).WaitAsync(30.Seconds());

  This frees the thread (removing the self-starvation), terminates both systems in parallel, and preserves the verify-shutdown semantics (WaitAsync throws TimeoutException if a system fails to stop in time). Same idiom already used by QuickRestartSpec and DistributedPubSubRestartSpec.

No production code is touched — this is a test-only change.

Verification

• Builds clean (Release, 0 warnings / 0 errors).
• Ran NodeChurnSpec locally 4 consecutive times — Passed! Failed: 0, Passed: 3 every run (~44–48s each).

Notes

This is one of two flaky-test fixes split out from the #8282 investigation. The CircuitBreakerSpec timing fragility (same family of cold-start/thread-pool root cause) will be addressed in a separate PR.

[Aaronontheweb]

Basically this addresses a ton of legacy sync-over-async issues from the original TestKit / MNTR designs