Discovery Memory managment doesn't scale

At present discovery doesn't stream asset objects from memory to disk as fragments. On large deployments this causes Discovery to consume large amounts of memory.

For larger deployments it would be preferred to have Asset objects stream to disk from memory and the memory returned. As this will allow a single instance to handle allot larger range of IPs on the same hardware and Memory assignment.

It will also prevent running out of memory errors some large deployments could experience while trying to size Discovery Instance Hosts.

If a thread could be dedicated to handle the streaming to disk and then the final rebuild into a single xml for upload, this would greatly improve Discovery's scalability.