DataNode: Decouple startup logic from Graylog server startup logic

Data Node startup logic contains a lot of duplicated code from Graylog startup classes.
While it makes sense that the startup mechanisms of Graylog should also be employed by data node, the duplication is a source for introducing errors when logic changes only on one side.
Startup should be unified, e.g. by providing a common code base for startups, which could also be reused e.g. in forwarders.