Oplog performance on primary: efficiency of uncommitedRecordIds

32 threads insert, 24 CPU box.
standalone: 1.15M insert/s
with oplog: 186K insert/s

Believed to be single threading in logOp from OplogHack mutex.
Contention from delete holding mutex whilst walking DLL is linear with list size.
Proposed fix of O(1) delete by retaining pointer handle to reference back