[fix][test]fix flaky test MLPendingAckStoreTest.testMainProcess (apache#18310)

Fixes apache#18230

### Motivation
`ManagedCursorImpl` has a known problem: when the `cursor.close` and the `cursor.switchLedger` are concurrent executing, a bad version exception is thrown. see: 

https://github.com/apache/pulsar/blob/0c3c175eb132d4ef0fb3a12842127dcb4fa1933d/managed-ledger/src/main/java/org/apache/bookkeeper/mledger/impl/ManagedCursorImpl.java#L2595-L2619

In this test, it hit this problem:

- `MlPendingAckStore.clearUselessLogData` will trigger `cursor.switchLedger` if it is the first execute `mard delete`.
- `MlPendingAckStore.close` will trigger `cursor.close`

### Modifications

if `MlPendingAckStore.close` fail, just retry, it will be ok

Author: poorbarcode

pulsar-broker/src/test/java/org/apache/pulsar/broker/transaction/pendingack/impl/MLPendingAckStoreTest.java

@@ -256,8 +256,24 @@ public Object answer(InvocationOnMock invocation) throws Throwable {
         Assert.assertTrue(mlPendingAckStoreForRead.pendingAckLogIndex.keySet().iterator().next().getEntryId() > 19);
 
         // cleanup.
-        mlPendingAckStoreForWrite.closeAsync().get();
-        mlPendingAckStoreForRead.closeAsync().get();
+        closePendingAckStoreWithRetry(mlPendingAckStoreForWrite);
+        closePendingAckStoreWithRetry(mlPendingAckStoreForRead);
+    }
+
+    /**
+     * Why should retry?
+     * Because when the cursor close and cursor switch ledger are concurrent executing, the bad version exception is
+     * thrown.
+     */
+    private void closePendingAckStoreWithRetry(MLPendingAckStore pendingAckStore){
+        Awaitility.await().until(() -> {
+            try {
+                pendingAckStore.closeAsync().get();
+                return true;
+            } catch (Exception ex){
+                return false;
+            }
+        });
     }
 
     /**