Don't tune eviction thread count when the count is fixed

During some performance workload investigation we found that the eviction worker thread tuning algorithm was causing disruption to throughput, even if the number of eviction threads was not dynamic (i.e: when eviction=(threads_max=X,threads_min=X) where X is the same for min and max.

We should not run the tuning process if it can't make a difference.