Add streaming variant of off heap to not create byte[] instances

Currently off heap has to create a byte[] to copy memory from off heap. This is a waste of CPU and memory to do so. Instead we should have a streaming variant Output that the GlobalMarshaller can use that doesn't need all of this unnecessary copying. Unfortunately GlobalMarshaller is currently internally hard coded to require the usage of byte[] so we would have to tweak some code there as well to allow this.