Avoid possible timeout initializing replications

If 2 different replications start and finish at about the same
time, there's a chance one is doing a synchronous gen_server
call to the replication manager (to notify it that it started)
while the replication manager is doing a call to the replication
supervisor because the other replication just finished.

COUCHDB-1461

src/couch_replicator/src/couch_replicator_manager.erl

@@ -431,7 +431,14 @@ replication_complete(DocId) ->
             % We want to be able to start the same replication but with
             % eventually different values for parameters that don't
             % contribute to its ID calculation.
-            _ = supervisor:delete_child(couch_replicator_job_sup, BaseId ++ Ext);
+            case erlang:system_info(otp_release) < "R14B02" of
+            true ->
+                spawn(fun() ->
+                    _ = supervisor:delete_child(couch_replicator_job_sup, BaseId ++ Ext)
+                end);
+            false ->
+                ok
+            end;
         #rep_state{} ->
             ok
         end,