Discard new writes without closing connection on AbortWritesEvent

[idelpivnitskiy]

Motivation:

Behavior of SslHandler in Netty 4.1.115 changed. Now it can wrap data and flush them to the network even if flush was not requested. In result, it affected our test ConnectionCloseHeaderHandlingTest.PipelinedRequestsTest.serverCloseSecondPipelinedRequestWriteAborted that we had to disable in #3097 to proceed with the upgrade.
Debugging showed that when we receive AbortWritesEvent, the following behavior of WriteStreamSubscriber depends on how many activeWrites we have. With 4.1.115 netty will flush data, there will be 0 activeWrites and it will close (with reset) the outbound. In result, we lose data of the previous response in the pipeline that we haven't read yet.

Modifications:

• Use listenerDiscard instead of channelClosed that will make sure new writes will be discarded without prematurely affecting the connection state regardless of the activeWrites count.
• Unskip the test.

Result:

Graceful handling of connection: close header does not depend on flushing behavior.

[chrisvest]

Sounds good from the description. I can't tell if the close outbound is delayed, or just doesn't happen, though.