Ticket 49696: replicated operations should be serialized
Bug: there was a scenario where two threads could process replication operations in parallel.
The reason was that for a new repl start request the repl conn flag is not set and the
connection is made readable.
When the start repl op is finished, the flagi set, but in a small window the supplier could
already have sent updates and more_data would trigger this thread also to continue to process
repl operations.
Fix: In the situation where a thread successfully processed a start repl request and just set the repl_conn
flag do not use more_data.
Reviewed by: Thierry, thanks