Restore functionality to existing MNTR specs

[Aaronontheweb]

I'm not sure what change caused this exactly (could have been an XUnit upgrade,) but since then all of the Akka.Remote and Akka.Cluster MNTR specs have thrown an error upon discovery and have not been run. The discovery error is not logged as a fault by the MNTR runner process itself, hence why it hasn't shown up at all in any of our integration tests.

Here's the error, from the MNTR logs on TeamCity:

[17:36:16] :		 [MultiNodeTests] System.ArgumentException: [Akka.DistributedData.Tests.MultiNode.DurableDataSpecNode1] or one of its base classes must specify constructor, which first parameter is a subclass of Akka.Remote.TestKit.MultiNodeConfig
[17:36:16] :		 [MultiNodeTests] Server stack trace: 
[17:36:16] :		 [MultiNodeTests]    at Akka.MultiNodeTestRunner.Discovery.FindConfigConstructor(Type configUser)
[17:36:16] :		 [MultiNodeTests]    at Akka.MultiNodeTestRunner.Discovery.RoleNames(Type specType)
[17:36:16] :		 [MultiNodeTests]    at Akka.MultiNodeTestRunner.Discovery.OnMessage(IMessageSinkMessage message)
[17:36:16] :		 [MultiNodeTests]    at Xunit.MessageSinkWithTypesAdapter.OnMessageWithTypes(IMessageSinkMessage message, HashSet`1 messageTypes)
[17:36:16] :		 [MultiNodeTests]    at System.Runtime.Remoting.Messaging.StackBuilderSink._PrivateProcessMessage(IntPtr md, Object[] args, Object server, Object[]& outArgs)
[17:36:16] :		 [MultiNodeTests]    at System.Runtime.Remoting.Messaging.StackBuilderSink.SyncProcessMessage(IMessage msg)

To fix this issue, the following libraries all need to be updated to use the new format (need default constructors for all specs) in order to be run again; either that or we need to fix the functionality of the MNTR discovery engine to be backwards compatible:

• Akka.DistributedData.Tests.MultiNode
• Akka.Cluster.Tests.MultiNode
• Akka.Remote.Tests.MultiNode

I don't feel comfortable releasing 1.3.3 until this is done so we can verify that the changes we've made to Akka.Cluster, Cluster.Sharding, and DistributedData are safe.

[Aaronontheweb]

Looks like the issue here is an edge-case with some specific tests that cause the test runner to crash, such as:

akka.net/src/core/Akka.Remote.Tests.MultiNode/RemoteReDeploymentSpec.cs

Lines 267 to 270 in 0045438

	public class RemoteReDeploymentSlowMultiNetSpec : RemoteReDeploymentSpec
	{
	public RemoteReDeploymentSlowMultiNetSpec():base(typeof(RemoteReDeploymentSlowMultiNetSpec))
	{ }

The LINQ being used here isn't able to find a valid match:

akka.net/src/core/Akka.MultiNodeTestRunner/Discovery.cs

Lines 110 to 123 in 0045438

	#if CORECLR
	var ctorWithConfig = configUser
	.GetConstructors(BindingFlags.NonPublic | BindingFlags.Public | BindingFlags.Instance)
	.FirstOrDefault(c => null != c.GetParameters().FirstOrDefault(p => p.ParameterType.GetTypeInfo().IsSubclassOf(baseConfigType)));
	current = current.GetTypeInfo().BaseType;
	#else
	var ctorWithConfig = configUser
	.GetConstructors(BindingFlags.NonPublic | BindingFlags.Public | BindingFlags.Instance)
	.FirstOrDefault(c => null != c.GetParameters().FirstOrDefault(p => p.ParameterType.IsSubclassOf(baseConfigType)));
	current = current.BaseType;
	#endif
	if (ctorWithConfig != null) return ctorWithConfig;

If we can fix the LINQ statements there or modify these specs to conform to the standard they're currently looking for, that should fix the issue.