New implementation of PublisherInputStream that improves performance and fixes race conditions

[spericas]

New implementation of PublisherInputStream, now renamed to DataChunkInputStream. This new implementation fixes a couple of important problems:

1. It implements the ability to read into byte arrays, not just one byte at a time for better performance and,
2. It changes the implementation to avoid a race condition when accessing the data chunks held in CompletableFuture's. The race condition resulted in certain tests to hang if a thread raced and updated the value of the old processed variable.

There is also a new internal document docs-internal/datachunkinputstream.md that describes the new implementation. Credit to Oleksandr Otenko.

Signed-off-by: Santiago Pericasgeertsen santiago.pericasgeertsen@oracle.com

[danielkec]

How about to rename it to something completely different as it is not a publisher nor subscriber.

public class FlowInputStream extends InputStream{
...
 private class DataChunkSubscriber implements Flow.Subscriber<DataChunk>{
     public void on...
     public void on...
     public void on...
     public void on...
     ...
    }
}

To avoid users of the api to accidentally violate rule §1.10 like this:

SubscriberInputStream subscriber = new SubscriberInputStream(publisher);
publisher.subscribe(subscriber);

[spericas]

I like the idea of the private class. I'll explore that.

[danielkec]

onSubscribe/onNext/onComplete/onError

[olotenko]

This change is focusing on correct synchronization (hand-over-hand) and reading in quantities larger than 1 byte. Looks good.