tcp: do not clear again skb->csum in tcp_init_nondata_skb()

tcp_init_nondata_skb() is fed with freshly allocated skbs.
They already have a cleared csum field, no need to clear it again.

This is based on Neal review on commit 3b11775 ("tcp: do not mangle
skb->cb[] in tcp_make_synack()"), noticing I did not clear skb->csum.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/tcp_output.c

@@ -381,7 +381,6 @@ static void tcp_ecn_send(struct sock *sk, struct sk_buff *skb,
 static void tcp_init_nondata_skb(struct sk_buff *skb, u32 seq, u8 flags)
 {
 	skb->ip_summed = CHECKSUM_PARTIAL;
-	skb->csum = 0;
 
 	TCP_SKB_CB(skb)->tcp_flags = flags;
 	TCP_SKB_CB(skb)->sacked = 0;