As it stands, when there is a bad syncset that can never be applied (without manual intervention by a user), the reconciler will hot loop until exponential backoff kicks in.  Ideally, we should be able to detect when a syncset is impossible to apply and avoid thrashing the reconciler with 10 or 15 loops (for each bad sync) before backoff starts. 

Examples of bad syncsets:

• syncing an object with an invalid/non-existent namespace
• syncing a object of a Kind that has no matching CRD on the target server

Particularly, the staging environment has a very large amount of these bad syncsets.