XDCR: Items stuck in replication queue after warmup, replicators awake, xdcr idle => updates lost

Same bug description as MB-11228 which fixed the UPR connection problem however, it now appears data is not completely replicated to destination cluster. More details below.

Build
-------
3.0.0-764

Scenario (internal repl = UPR, XDCR on UPR)
-------------
1. Uni-xdcr between 2 clusters
2. Load 1000 items on source
3. Warmup default bucket or one of the source nodes
4. Delete last 30% of keys loaded (loadOne699 to loadOne999)
5. Do item count, revID, value verification on both clusters.

Testrunner command to run test
--------------------------------------------------
./testrunner -i /tmp/XDCR-sanity.ini -t xdcr.uniXDCR.unidirectional.load_with_async_ops_with_warmup,items=1000,rdirection=unidirection,ctopology=chain,doc-ops=delete,warm=source,replication_type=xmem

Observations
--------------------

• 316 items stuck in outbound replications queue on .187(warmed up node)
• 16 replicators awake
• replication idle

Source: 700 items
Destination : 789 items

Live cluster (with xdcr trace logging enabled)
-------------------
Source - http://10.3.4.186:8091/
Destination - http://10.3.4.188:8091/

Why test blocker?
-------------------------
All xdcr tests with warmup fail