Don't count bucket allocation failures for UMA zones as their own failures.
There are good reasons for this to happen, such as recursion prevention, etc.
and they are not fatal since buckets are just an optimization mechanism.
Real bucket allocation failures are any way counted by the bucket zones
themselves, and we don't need double accounting there.