When Server.Stop() returns it is not guaranteed that the loopyWriter
go-routine terminated.

This can lead to a panic or a testing.(*common).logDepth() race
condition in Go Tests because t.Log is used after or during the testcase
terminates.
This can happen when:
- a GRPC server is started in a Go test,
- the GRPC logs are forwarded to t.Log,
- loopyWriter.run logs an error after server.Stop() and the Test
  method returns.

grpc@v1.57.0/internal/leakcheck is unable to detect that the
loopyWriter go-routine continues to run after server.Stop() because it
waits up to 10s for go-routines to terminate after a test finishes.
The loopyWriter returns much faster after Stop() returns.

To make server.Stop() wait until loopyWriter terminated:
- rename the existing writerDone field, which is only used in tests, to
  loopyWriterDone, the writerDone channel is closed when the loopyWriter
  go-routine exited
- change http2server.Close to wait until loopyWriterDone is closed

…inated"

This reverts commit ad96f3c.

It was not guaranteed that when Stop() and GracefulStop() returned, the
loopyWriter go-routines terminated.
This can lead to a panic or a testing.(*common).logDepth() race
condition in Go Tests because t.Log is used after or during the testcase
terminates.
This can happen when:
- a GRPC server is started in a Go test,
- the GRPC logs are forwarded to t.Log,
- loopyWriter.run logs an error after server.Stop() and the Test
  method returns.

grpc@v1.57.0/internal/leakcheck is unable to detect that the
loopyWriter go-routine continues to run after server.Stop() because it
waits up to 10s for go-routines to terminate after a test finishes.
The loopyWriter returns much faster after Stop() returns.

To make ensure Stop and GracefulStop only return after the loopyWriter
go-routine terminated:
- rename http2Server.writerDone to loopyWriterDone, the channel was only
  used in test and is closed when loopyWriter returns
- wait in http2Server.HandleStreams() until loopyWriterDone was closed
- wait in Server.Stop() until all connections were removed, this is
  already done in GracefulStop.
  To abort GracefulStop() when Stop() is called a local copy of Server.lis and
  Server.conns was created and both fields were set to nil and then
  s.csv.Broadcast() was called to wake up an eventually sleeping
  GracefulStop() execution. This does not work anymore when we need to
  wait for connections to be removed in Stop() we need to check the
  length of Server.conns.
  Instead a new abortGracefulStop atomicBool is introduced, that is
  set to true to indicate that a GracefulStop() invocation should be
  aborted when Stop() is run.

…utine_stop

…utine_stop

The call was only done in Stop() and missing in GracefulStop()

Move code that is used in Stop and GracefulStop to helper functions
instead of duplicating it. This makes it easier to compare Stop() and
GracefulStop.

…utine_stop

…utine_stop

…raining"

This reverts commit 0264c98ecd860431ac82a4e799cf6282e08e8a04.

- drain transporters after waiting serve handlers to terminate
- call closeServerTransports only on non graceful shutdowns, to abruptly
  close the connections

…utine_stop

Add 2 testcases that ensure that:
- GracefulStop blocks and waits until ongoing client grpc calls finis,
- Stop() returns despite ongoing blocking grpc calls and clients receive
  a connection error

The file was added accidentally