Reduce amount of metadata sent around.

Instead of using int32 for tracking dataset types for leaf node, we now use int8. This reduces that size of data being gathered to root node.