Rudder

The query to get last nodes runs configuration is ok, but when we have a significant number of nodes to filter on, it gets pretty slow (from 50ms to 2s with 900 nodes)
There is a threashold where it is cheaper to simply take all nodes, and filter in scala (as taking all nodes is more about 1s on my test system)

(note: this is also true in 3.1 implementation to get expected reports)

This needs more testing, obviously, but it could improve dashboard loading at web interface start, or when all nodes are answering at the same time