SocketConnector is now a Callable, all tests pass

LOGBACK-848: refactored SocketAppenderBase

invoke setRunner method in ConcurrentServerRunner so that InstrumentedConcurrentServerRunner's overriden impl can be used in tests

Since the appender now has some an asynchronous task, this is a best
practice.  Moreover, some test runners may not be able to terminate
properly if there are non-daemon threads remaining in the JVM.

Previously, we exited only when interrupted.  But if the executor is
shutting down, there's no reason for the receiver to try to continue
running.

This was a left over from when the receiver had its own executor
service, and ExecutorService.shutdownNow was used to shut it down on
stop().

Previously, exited only on interrupt, but if the executor is shutting
down, there's no reason for us to wait to be interrupted.

ExecutorService.shutdownNow returns the list of jobs that never
executed, not the list of jobs that didn't terminate.

Conflicts:
	logback-core/src/main/java/ch/qos/logback/core/ContextBase.java

Mutable fields in SocketAppenderBase were previously exposed as
protected fields and modified by constructors in concrete subclasses.
This commit improves encapsulation of the base class by exposing
protected constructors for subclasses to invoke rather than directly
exposing the fields.  These new constructors are marked as deprecated,
as the convenience constructors on the base classes (which are also
deprecated) are the sole reason for their existence.  A future release
should remove all but the no-arg constructor from the base class and
subclasses.

SocketAppenderBase now uses SocketConnector for its (re-)connection
logic, and uses an asynchronous task to dispatch events to the remote
receiver.  A configurable queue is utilized to relay logging events from
the append method to the dispatch task.  When the queue length is zero
(the default), a SynchronousQueue is utilized, preserving the previous
appender behavior.  When the queue length is greater than zero, a
bounded queue is utilized, allowing the appender to efficiently drop
logging events when remote receiver (or network) cannot keep up with the
rate of logging events delivered to the appender.