Logging of cluster rpc method exceptions at warn level is incorrect

The HAPartitionImpl handler for cluster rpc method invocations logs Exceptions and Errors thrown from the reflected method invocation at warn level. This results in spurious warnings when the underlying invocation throws an exception as part of its normal operation. An example of where this occurs is when an ha-sigleton binds a jndi entry into its node, and a client does a lookup of the binding through ha-jndi. In a 3 node cluster A, B, C with the ha-singleton on A, a ha-jndi request going to B will produce the following NameNotFoundException on C:

06:52:07,359 WARN [DefaultPartition] javax.naming.NameNotFoundException: TheSingleton not bound

These exceptions should be logged at trace level.