XDCR between ephemeral buckets gets stuck

Yet another issue when ephemeral buckets hang.
The root cause is probably the same but I still need this ticket for tracking purposes.

Setup:

• 1 node -> 1 node replication
• 1 bucket per cluster (no eviction)
• 100M documents on the source side
• 4 source nozzles, 64 target nozzles

Replication gets stuck after sending out 3904280 documents.

       "docs_processed": 4053434,

        "docs_received_from_dcp": 3904280,

        "docs_written": 3904280,