Race condition - possible to get null when reading instance config while adding/removing instance

[xyuanlu]

Describe the bug

When adding/removing instance from a cluster, it is possible that at a point of time, instance config is gone but the INSTANCE ZNode is still there. This would cause config map in helix controller cache to have instance mapped to null config. We need to add NPR guard in pipeline to prevent transient pipeline failure.

The failure is not a blocking error causing further pipeline error but we would like to avoid noise in log.