Fix flaky ResourceAwareTasksTests (opensearch-project#20863)

Race condition between request completion and task resource tracking
cleanup.

The sequence of events:
1. Task is cancelled via `CancelTasksRequest`
2. The node operation throws `TaskCancelledException`
3. The response is sent back to the caller, which counts down
   `requestCompleteLatch`
4. The test's main thread wakes up from `requestCompleteLatch.await()`
   and asserts `resourceTasks.size() == 0`
5. Meanwhile, `TaskResourceTrackingService.stopTracking()` (which
   calls `resourceAwareTasks.remove()`) is invoked asynchronously
   via a `resourceTrackingCompletionListener` registered in
   `TaskManager.register()`

Steps 4 and 5 race. I was able to reproduce the failure locally using
`stess-ng` and verify this fix.

Signed-off-by: Andrew Ross <andrross@amazon.com>

Author: andrross

server/src/test/java/org/opensearch/action/admin/cluster/node/tasks/ResourceAwareTasksTests.java

@@ -410,7 +410,9 @@ public void onFailure(Exception e) {
         // Waiting for whole request to complete and return successfully till client
         taskTestContext.requestCompleteLatch.await();
 
-        assertEquals(0, resourceTasks.size());
+        // The task may not be removed from resourceAwareTasks immediately after the request completes
+        // because stopTracking is called asynchronously via the task's resource tracking completion listener.
+        assertBusy(() -> assertEquals(0, resourceTasks.size()));
         assertNull(throwableReference.get());
         assertNotNull(responseReference.get());
         assertEquals(1, responseReference.get().failureCount());