TCP engine: Fix a race condition related to con lifetime management

Connections which received a read event concurrently with their
expiration time could trigger an indefinitely long, CPU wasteful, rapid
succession of back-forth passing of the connection object between TCP Main
and a TCP worker process.

The fix involves adding a TCP connection lifetime update in the TCP worker
just after receiving the connection object from TCP Main (this is correct
- there is some pending data, the connection is non-idle), before the
expiration check is done.

Thanks to Bogdan for helping with troubleshooting

(cherry picked from commit 9918171)

net/net_tcp_proc.c

@@ -168,6 +168,8 @@ inline static int handle_io(struct fd_map* fm, int idx,int event_type)
 				 * handle_io might decide to del. the new connection =>
 				 * must be in the list */
 				tcpconn_listadd(tcp_conn_lst, con, c_next, c_prev);
+				/* pending event on a connection -> prevent premature expiry */
+				tcp_conn_set_lifetime(con, tcp_con_lifetime);
 				con->timeout = con->lifetime;
 				if (reactor_add_reader( s, F_TCPCONN, RCT_PRIO_NET, con )<0) {
 					LM_CRIT("failed to add new socket to the fd list\n");