Multi-flavor Slack ClusterQueue

[dgrove-oss]

The slack cluster queue lending limit adjustment should be flavor aware. The initial implementation implicitly assumes that all nodes in the cluster belong to the first/only flavor. There are two main pieces to implement this feature.

1. The node monitoring code should be extended to maintain per-flavor lists of NoSchedule resources.
2. The slack cluster queue lending limit adjust logic would use the extended information to update the lending limit for every flavor instead of hardcoding flavor[0].