To improve the device watching mechanism a fetch all after time interval method should be implemented

Description

After several discussions the idea to improve the watching mechanism is to remove the stream on update or sample and replace it with some kind of polling mechanism. The controller then starts the process in a timed interval to collect the whole dataset available in the YANG model and writes it to the database. This is supposed to deal with inconsistency and scalability issues.

Alternative (probably not suitable) solution: Make use of allow_aggregation field in gNMI subscribe options. This way, there are still issues with inconsistency but it already helps with the problems. So, maybe this could be used in the future.

Related to #256 (closed)