test(coverage): close Theme 8 — 13 test-coverage findings, +35 tests

13 well-bounded test-coverage gaps closed across 11 test projects.
Net +35 regression tests; no production code changes except the
SiteEventLogger src reference unchanged (W3 redacted only test code).

Test additions:
- CLI-022: CommandTreeTests pinned-count assertion bumped 14→16 and
  3 InlineData rows added for the audit + bundle command groups.
- Commons-020: new TransportRecordsTests covers BundleManifest /
  ExportSelection / ImportPreview / ImportResolution / ImportResult —
  ctor + System.Text.Json round-trip + record-equality (14 tests).
- CD-024: SPLIT-RANGE failure-continuation now under
  EnsureLookahead_SecondSplitThrows_LoopAborts_FirstBoundaryStillCommitted
  (Skippable MS-SQL fixture); production-shape rowversion delete
  asserted by DeleteDeploymentRecord_CurrentRowVersion_StubAttachPath_DeleteSucceeds.
- CentralUI-033: new QueryStringDrillInTests with 4 bUnit cases for
  Transport + SiteCalls drill-in / query-string handling.
- DM-024: probe actors (ReconcileProbeActor, SerializationProbeActor,
  ArtifactProbeActor) refactored from static fields to per-test instances
  (Interlocked on counter) — all 31 callers updated; no production
  changes required.
- HM-022: real-time PeriodicTimer test flake fixed by replacing
  fixed-budget Task.Delay with a RunLoopUntil poll-until-condition
  helper (5s/25ms). Production loop untouched.
- InboundAPI-023: new EndpointExtensionsTests covers the
  POST /api/{methodName} composition wiring via TestServer (7 cases:
  happy path, missing key 401, unknown method 403, invalid JSON 400,
  missing param 400, script-throws 500 sanitised, AuditActorItemKey
  stash invariant).
- MgmtSvc-021: 6 new ManagementActorTests cover the Transport bundle
  handlers (role gate for Export/Preview/Import, unknown-name
  ManagementCommandException, blocker-rejection, dedupe last-write-wins).
- SCA-006: SiteCallQueryRequest_StuckOnly_CursorAtNonStuckBoundary_SkipsToNextStuckRow
  pins the missing boundary case.
- SEL-023: stress-test `bool stop` promoted to `volatile bool` for
  cross-thread visibility under release/JIT.

Verify-only resolutions:
- NS-024: closed by NS-019 (commit ac96b83 deletion of
  NotificationDeliveryService + its test file). No edits needed.
- NotifOutbox-008: FallbackMaxRetries/FallbackRetryDelay are private
  forward-compat constants returned only when no SMTP-config row exists
  (in which case EmailNotificationDeliveryAdapter returns Permanent,
  bypassing the values entirely). Marked Resolved with note.
- Transport-010: Overwrite child-collection sync covered by the T-001/
  T-002 tests added in commit e3ca9af; per-IP throttle by
  BundleUnlockRateLimiterTests; failed-session retention by
  BundleSessionStoreTests; T-009 closed structurally via AsyncLocal.
  Marked Resolved by reference.

Build clean; all 11 affected test suites green. README regenerated:
33 open (was 46).

tests/ScadaLink.SiteEventLogging.Tests/EventLogPurgeServiceTests.cs

@@ -9,6 +9,15 @@ public class EventLogPurgeServiceTests : IDisposable
     private readonly string _dbPath;
     private readonly SiteEventLogOptions _options;
 
+    /// <summary>
+    /// SiteEventLogging-023: stop flag for the concurrent stress test. Declared as
+    /// a <c>volatile</c> field so every writer thread observes the main thread's
+    /// `_stop = true` write without depending on JIT/runtime quirks. A plain
+    /// <c>bool</c> local would be legal-cached in a register inside the tight
+    /// <c>while (!_stop)</c> loop under release-mode optimisation.
+    /// </summary>
+    private volatile bool _stop;
+
     public EventLogPurgeServiceTests()
     {
         _dbPath = Path.Combine(Path.GetTempPath(), $"test_purge_{Guid.NewGuid()}.db");
@@ -282,7 +291,13 @@ public class EventLogPurgeServiceTests : IDisposable
         };
 
         var exceptions = new System.Collections.Concurrent.ConcurrentBag<Exception>();
-        var stop = false;
+        // SiteEventLogging-023: must be volatile so the writer threads observe the
+        // main thread's `stop = true` flip after the purge task completes. Without
+        // it, a release-mode JIT is permitted to cache the `stop = false` read in
+        // a register inside the tight `while (!stop)` loop and never see the flip,
+        // causing the writer tasks to hang past xUnit's per-test timeout instead
+        // of asserting `Empty(exceptions)`.
+        _stop = false;
 
         var purgeTask = Task.Run(() =>
         {
@@ -298,7 +313,7 @@ public class EventLogPurgeServiceTests : IDisposable
         {
             try
             {
-                while (!stop)
+                while (!_stop)
                 {
                     await _eventLogger.LogEventAsync("script", "Info", null, "Concurrent", "Concurrent write");
                 }
@@ -307,7 +322,7 @@ public class EventLogPurgeServiceTests : IDisposable
         })).ToArray();
 
         await purgeTask;
-        stop = true;
+        _stop = true;
         await Task.WhenAll(writeTasks);
 
         Assert.Empty(exceptions);