[Proposal] min nodes should be dynamic

[mattbennett]

I installed this is a cluster with a taint set on the preemptible nodes, and only certain pods configured with a toleration.

With the default setting of 0 for NODE_POOL_FROM_MIN_NODE, the shifter was constantly adding nodes to the preemptible pool, even though no additional workload could be run there. The node pool autoscaler was then eventually removing them again because the pool was over capacity.

I stopped the thrashing by increasing NODE_POOL_FROM_MIN_NODE until the pods without the toleration all fit, but this is an unfortunate bit of manual tuning.

It would be much nicer if the shifter determined whether to take action by inspecting whether there were any pods running on the "from" pool that could run on the "to" pool.