Flaky Test from NativeMemoryAllocationTests

[VijayanB]

Flaky test is coming from https://github.com/opensearch-project/k-NN/actions/runs/11113230012/job/30877097217?pr=2167

Tests with failures:
888 tests completed, 3 failed, 223 skipped
 - org.opensearch.knn.index.memory.NativeMemoryAllocationTests.testIndexAllocation_closeBlocking
 - org.opensearch.knn.index.memory.NativeMemoryAllocationTests.classMethod

 REPRODUCE WITH: gradlew ':test' --tests "org.opensearch.knn.index.memory.NativeMemoryAllocationTests" -Dtests.seed=9A46305B6CCE0712 -Dtests.security.manager=false -Dtests.locale=en-US -Dtests.timezone=Etc/UTC -Druntime.java=21

 gradlew ':test' --tests "org.opensearch.knn.index.memory.NativeMemoryAllocationTests" -Dtests.seed=9A46305B6CCE0712 -Dtests.security.manager=false -Dtests.locale=en-US -Dtests.timezone=Etc/UTC -Druntime.java=21

 gradlew ':test' --tests "org.opensearch.knn.index.memory.NativeMemoryAllocationTests.testIndexAllocation_closeBlocking" -Dtests.seed=9A46305B6CCE0712 -Dtests.security.manager=false -Dtests.locale=tt -Dtests.timezone=Europe/Ljubljana -Druntime.java=21

[0ctopus13prime]

This PR(#2139) is trying to fix this flaky testing.
In which, it changed its way to read a lock first then followed by calling close().

    public void testIndexAllocation_closeBlocking() throws InterruptedException, ExecutionException {
        // Prepare mocking and a thread pool.
        WatcherHandle<FileWatcher> watcherHandle = (WatcherHandle<FileWatcher>) mock(WatcherHandle.class);
        ExecutorService executorService = Executors.newSingleThreadExecutor();

        // Enable `KNN_FORCE_EVICT_CACHE_ENABLED_SETTING` to force it to block other threads.
        // Having it false will make `IndexAllocation` to run close logic in a different thread.
        when(clusterSettings.get(KNN_FORCE_EVICT_CACHE_ENABLED_SETTING)).thenReturn(true);
        NativeMemoryAllocation.IndexAllocation blockingIndexAllocation = new NativeMemoryAllocation.IndexAllocation(
            mock(ExecutorService.class),
            0,
            0,
            null,
            "test",
            "test",
            watcherHandle
        );

        // Acquire a read lock
        blockingIndexAllocation.readLock();

        // This should be blocked as a read lock is still being held.
        Future<?> closingThread = executorService.submit(blockingIndexAllocation::close);

        // Check if thread is currently blocked
        try {
            closingThread.get(5, TimeUnit.SECONDS);
            fail("Closing should be blocked. We are still holding a read lock.");
        } catch (TimeoutException ignored) {}

        // Now, we unlock a read lock.
        blockingIndexAllocation.readUnlock();
        // As we don't hold any locking, the closing thread can now good to acquire a write lock.
        closingThread.get();

        // Waits until close
        assertTrue(blockingIndexAllocation.isClosed());
        executorService.shutdown();
    }

[0ctopus13prime]

PR(#2139) fixed the timing issue.
We can close this and reopen issue if it is still reoccuring.
But at least, in the PR, the failed testing now has passed.