SyncWriter.flush() now throw Exception when previous task was failed

[ekuvardin]

Problem connected to this one #475

What's the cause?

A problem when encoding/decoding (or something else) throws an error.

The user writes

WriterSettings settings = WriterSettings.newBuilder()
                .setTopicPath(topicName)
                .setCodec(10000) // or .setCodec(Codec.CUSTOM) for new version
                .build();
        SyncWriter writer = client.createSyncWriter(settings);
         writer.initAndWait();


        for (byte[] testMessage : testMessages) {
            writer.send(Message.newBuilder().setData(testMessage).build());
        }

       writer.flush();
       writer.shutdown(1, TimeUnit.MINUTES);

Code in writer.send failed because codec CUSTOM did not exist.
But the problem is when it's failed before or writer.flush();

When we call writer.flush(), we finally end up in flushImpl.
Take a look at in WriterImpl

this.lastAcceptedMessageFuture.isDone()
                    ? CompletableFuture.completedFuture(null)
                    : this.lastAcceptedMessageFuture.thenApply(v -> null);

If the thread was too quick, then lastAcceptedMessageFuture - DONE(with error inside future), we return CompletableFuture.completedFuture(null) and writer.flush() doesn't fail.

If the thread is slow, then lastAcceptedMessageFuture is NOT COMPLETED, and we call lastAcceptedMessageFuture.thenApply(v -> null) and failed with RuntimeException wraped by CompletionException

Behavior depends on the speed of the test/machine and so on.

Solution
I think that we should throw an error in writer.flush()
Return his. lastAcceptedMessageFuture.thenApply(v -> null); despite whether lastAcceptedMessageFuture is done or not

Test

A bit confused where to write the test.
Test depends on #447 (Codec class)

Also, in MR below, we can uncomment Ignore test in YdbTopicsCodecIntegrationTest they will pass.
Maybe wait for MR to push?
On the other hand I can write simple one test -> just push Thread.sleep(5000) before writer.flush();

[pnv1]

PR 447 is pushed, but in next release branch, not in master. Test that depends on new codecs should probably also be added to new release branch