Performance regression in "Delta recovery after graceful failover" (reb_recovery_100M_dgm_kv_delta_graceful)

Between builds 5.5.0-1911 and 5.5.0-1979 the performance of this test regressed from 3.7 mins to 52.3 mins. It actually regressed in a couple of steps:

Build 1979 includes the new statistical LRU counter. I've re-run build 1979 with the eviction policy reverted to the old (ht_eviction_policy=2-bit_lru) - see http://perf.jenkins.couchbase.com/job/hestia/2344 - and this recovers most (but not all) of the regression - back to 9.3mins.

Therefore the bulk of this slowdown appears to be caused by the LRU change.