Reduce amount of wasted work on the server side

[jotelha]

9ca9021 results in an immense amount of wasted work on the server side in case of many objects in a dataset.

Possible solutions:

• Again only look out for certain objects. Without any guarantee of the order objects are written and reported via notifications, this bears the risk of missing out datasets.
• Instead of trying to register the dataset for every object, queue all notifications for a dataset and only try to register a certain grace period after the last notification for a specific dataset has arrived. This could be realized by a simple per-(UUID, base URI) timer that starts running at the first notification for a specific (UUID, base URI), is reset for every notification for the same (UUID, base URI), and triggers a registration attempt after no other notification for this (UUID, base URI) tuple has been received for a couple of seconds.