Problem statement:
Some configuration changes cause rolling restarts of the cluster (e.g. audit filter, enabling x509 auth). It is not immediately obvious which changes will cause restarts and for a big cluster, this is a huge hit to performance due to cold caches.

Ideas:
1. Block planner changes which would cause restarts until maintenance window (maybe with an opt-in for this, + notification that the config change is blocked)
2. Throttle replica set restarts, allowing time for the cache to be warmed on secondaries (would require mirrored reads in 4.4 to be effective)
3. Properly document what changes will cause rolling restarts