Add somekind of caching mechanism to ActionProcessingPipeline for jBPM to improve performance.

Each time an ESB service is invoked from a jBPM node a jUDDI lookup is made to find the endpoint reference for JBpmCallback Service. This reference will be used as the target to which the response will be sent.

The JBpmCallback Service is represented inside the message as LogicalEPR and for each LogicalEPR a new ServiceInvoker is created. This is the place where jUDDI is accessed. Each such jUDDI lookup takes about 200-300 ms.