Skip evaluating system jobs when nodes go down or are deregistered

[schmichael]

Continuation of #12955 (review)

As of Nomad 1.3.0 we use the same code for emitting node-update evaluations regardless of whether the node in question is going down (or deregistering) or coming back up.

When a node comes up, we re-evaluate all system jobs to ensure any relevant system jobs are started on the node.

However, there's no need to also re-evaluate all system jobs when a node goes down.

This could save a significant number of evaluations in clusters with system jobs and flappy nodes or a large network partition.

There may be opportunities to optimize the case where nodes are coming up, but I'm unsure of any at this time.