AutoReprepare verification can easily take 20% Q1 throughput

Observed on iris - when autoreprepare kicks in throughput goes from 119k/s to 95k/s.
This is down to the fact that auto reprepare stores a pointer to namespaces and indexers, but if the namespace or indexer gets discarded and rebuilt, autoreprepare will never be able to match namespaces / indexers and therefore amend version counters.
This will trigger full verification at each prepared execution, piling up on the namespace / indexer mutex.