Diskless replication: set / reset socket send timeout.

We need to avoid that a child -> slaves transfer can continue forever.
We use the same timeout used as global replication timeout, which is
documented to also affect I/O operations during bulk transfers.

src/rdb.c

@@ -1349,6 +1349,7 @@ void backgroundSaveDoneHandlerSocket(int exitcode, int bysignal) {
                 slave->id);
                 /* Restore the socket as non-blocking. */
                 anetNonBlock(NULL,slave->fd);
+                anetSendTimeout(NULL,slave->fd,0);
             }
         }
     }
@@ -1414,6 +1415,7 @@ int rdbSaveToSlavesSockets(void) {
              * We'll restore it when the children returns (since duped socket
              * will share the O_NONBLOCK attribute with the parent). */
             anetBlock(NULL,slave->fd);
+            anetSendTimeout(NULL,slave->fd,server.repl_timeout*1000);
         }
     }