Fix IllegalStateException in HttpChannelState.completed() after abort()

[lorban]

Fixes #12520

The following stack trace is logged when a certain sequence of events is happening in the FSM:

java.lang.IllegalStateException: s=HANDLING rs=COMPLETED os=ABORTED is=IDLE awp=false se=false i=false al=0
	at Jenkins Main ClassLoader//org.eclipse.jetty.ee9.nested.HttpChannelState.completed(HttpChannelState.java:1025)
	at Jenkins Main ClassLoader//org.eclipse.jetty.ee9.nested.HttpChannel.lambda$handle$0(HttpChannel.java:634)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.Callback$3.succeeded(Callback.java:170)
	at Jenkins Main ClassLoader//org.eclipse.jetty.ee9.nested.HttpOutput.complete(HttpOutput.java:493)
	at Jenkins Main ClassLoader//org.eclipse.jetty.ee9.nested.Response.completeOutput(Response.java:990)
	at Jenkins Main ClassLoader//org.eclipse.jetty.ee9.nested.HttpChannel.handle(HttpChannel.java:634)
	at Jenkins Main ClassLoader//org.eclipse.jetty.ee9.nested.ContextHandler$CoreContextHandler$CoreToNestedHandler.handle(ContextHandler.java:2873)
	at Jenkins Main ClassLoader//org.eclipse.jetty.server.handler.ContextHandler.handle(ContextHandler.java:1060)
	at Jenkins Main ClassLoader//org.eclipse.jetty.server.Server.handle(Server.java:182)
	at Jenkins Main ClassLoader//org.eclipse.jetty.server.internal.HttpChannelState$HandlerInvoker.run(HttpChannelState.java:662)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.Invocable$ReadyTask.run(Invocable.java:168)
	at Jenkins Main ClassLoader//org.eclipse.jetty.http2.server.internal.HttpStreamOverHTTP2$1.run(HttpStreamOverHTTP2.java:135)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.strategy.AdaptiveExecutionStrategy.runTask(AdaptiveExecutionStrategy.java:478)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.strategy.AdaptiveExecutionStrategy.consumeTask(AdaptiveExecutionStrategy.java:441)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.strategy.AdaptiveExecutionStrategy.tryProduce(AdaptiveExecutionStrategy.java:293)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.strategy.AdaptiveExecutionStrategy.run(AdaptiveExecutionStrategy.java:201)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.ReservedThreadExecutor$ReservedThread.run(ReservedThreadExecutor.java:311)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.QueuedThreadPool.runJob(QueuedThreadPool.java:979)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.QueuedThreadPool$Runner.doRunJob(QueuedThreadPool.java:1209)
	at Jenkins Main ClassLoader//org.eclipse.jetty.util.thread.QueuedThreadPool$Runner.run(QueuedThreadPool.java:1164)
	at java.base/java.lang.Thread.run(Thread.java:1583)

These are the calls and states that triggers this bug:

1. abort() / s=HANDLING rs=COMPLETING os=ABORTED is=IDLE awp=false se=false i=false al=0 - failure is java.io.IOException: Insufficient content written
2. completed() / s=HANDLING rs=COMPLETED os=ABORTED is=IDLE awp=false se=false i=false al=0 - throws IllegalStateException: s=HANDLING rs=COMPLETED os=ABORTED is=IDLE awp=false se=false i=false al=0

abort() incorrectly moves the request state to COMPLETED when it changes the state to WOKEN: this transition requires a redispatch that eventually calls completed() while the request state already is at its terminal position.

See #11210 as the same problem was fixed in EE10, but the fact that EE9 might have a similiar problem was blurred by the fact that the original issue had to do with GZIP which was buggy too.

[gregw]

Would be good to have a test for this use-case.