lightingd: fix memleak when we reload from db.

We always arm the funding_lockin_cb, even if we don't need to.  If we
have an short_channel_id already from the db, this was replacing it
and leaking the old one.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

lightningd/peer_control.c

@@ -1502,10 +1502,13 @@ static enum watch_result funding_lockin_cb(struct peer *peer,
 
 	loc = locate_tx(peer, peer->ld->topology, &txid);
 
-	peer->scid = tal(peer, struct short_channel_id);
-	peer->scid->blocknum = loc->blkheight;
-	peer->scid->txnum = loc->index;
-	peer->scid->outnum = peer->funding_outnum;
+	/* If we restart, we could already have peer->scid from database */
+	if (!peer->scid) {
+		peer->scid = tal(peer, struct short_channel_id);
+		peer->scid->blocknum = loc->blkheight;
+		peer->scid->txnum = loc->index;
+		peer->scid->outnum = peer->funding_outnum;
+	}
 	tal_free(loc);
 
 	/* In theory, it could have been buried before we got back