Fix crash in bogus OOM handling while TCP async in enabled.

The async support assums the c->async exists all the time, so error if we cannot alloc the c->async struct while creating a new TCP conn

(cherry picked from commit 0b270ad)

net/net_tcp.c

@@ -898,8 +898,10 @@ static struct tcp_connection* tcpconn_new(int sock, union sockaddr_union* su,
 			c->async->allocated = protos[si->proto].net.async_chunks;
 			c->async->oldest = 0;
 			c->async->pending = 0;
-		} else
-			LM_WARN("could not allocate async data for con!\n");
+		} else {
+			LM_ERR("could not allocate async data for con!\n");
+			goto error;
+		}
 	}
 
 	tcp_connections_no++;

net/proto_tcp/proto_tcp.c

@@ -217,8 +217,13 @@ static int proto_tcp_init(struct proto_info *pi)
 	}
 
 	/* without async support, there is nothing to init/clean per conn */
-	if (tcp_async!=0)
+	if (tcp_async!=0) {
+		/* be sure the settings are consistent, like having a minimum 2 value
+		 * if the tcp_async is enbled */
+		if (tcp_async_max_postponed_chunks<=1)
+			tcp_async_max_postponed_chunks = 2;
 		pi->net.async_chunks= tcp_async_max_postponed_chunks;
+	}
 
 	return 0;
 }