Abysmal dcp_xdcr drain rate(~100 times slower) compared to dcp_replica drain rate for replication with trace/debug logging

Build

4.0.0-2006

Found during manual testing. Pls see screenshot.

1. C1: .186,.187 C2: .188,.189, create default bucket. Start loading 10M onto .186.
2. After 300K keys are loaded, create replication C1 --> C2 with logLevel:Trace. Continue loading.
3. Watch dcp drain rate for inter and intra cluster replication.

Seeing intra cluster replication drain rate at 8-9K/sec and xdcr at ~80-100/sec

Consistently reproducible with logLevel: Trace/Debug.

Attaching logs from C1. Assigning to Xiaomei first.