Fix flaky test: SimplePublisherTest#stochastic_onNext_multiProducerSeemsThreadSafe (#4206)

Before this change, a producer in the test would not wait for all other producers to finish sending their messages before calling complete(). Because a producer could end up calling complete() before the other publishers finished sending, the test would see those "trailing" messages after the complete() as having been dropped. To fix this, we now wait to call complete() until all producers have finished writing their messages.

Author: millems

utils/src/test/java/software/amazon/awssdk/utils/async/SimplePublisherTest.java

@@ -414,6 +414,7 @@ private void seemsThreadSafeWithProducerCount(int producerCount) {
             AtomicLong messageSendCount = new AtomicLong(0);
             AtomicLong messageReceiveCount = new AtomicLong(0);
 
+            CountDownLatch producersDone = new CountDownLatch(producerCount);
             Semaphore productionLimiter = new Semaphore(101);
             Semaphore requestLimiter = new Semaphore(57);
             ExecutorService executor = Executors.newFixedThreadPool(2 + producerCount);
@@ -431,6 +432,10 @@ private void seemsThreadSafeWithProducerCount(int producerCount) {
                         productionLimiter.acquire();
                         publisher.send(messageSendCount.getAndIncrement());
                     }
+
+                    // Complete once all producers are done
+                    producersDone.countDown();
+                    producersDone.await();
                     publisher.complete().thenRun(() -> completed.complete(null)); // All but one producer sending this will fail.
                     return null;
                 }));