fix: wrong GTP length when Sequence Number is enabled and QoS is disa…

…bled

## Bug description
If you create a configuration that make a new GTP packet with Sequence Number
but you don't provide a QER IE (which is an optional IE), the resulting GTP
packet's length field is off by 4.

This make the packet total payload is shorter than indicated in the
header, resulting in interoperability issues (in particular, when
attempting to use a custom SMF that will push a different set of PFCP
rules than Free5GC's one, and where QoS management is not yet implemented).

For exemple, when creating a gNB using [go-gtp](https://github.com/wmnsk/go-gtp),
such packet is considered malformed, and is always dropped.

## Bug cause
This bug results from the following:
```c
ext_pdu_sess_ctrl_t *ext_pdu_sess;
// ...
if (opt_flag) {
    // ...
    payload_len += ((sizeof(*gtp1opt) + sizeof(*ext_pdu_sess));
}
```

When `ext_pdu_sess` is not initialized, `sizeof(*ext_pdu_sess)` returns `4` and not `0`.

## Bug mitigation
The fix is to split the incrementation in 2 steps:
1. increment by `sizeof(*gtp1opt)` if an option (Sequence Number, or Extension Header) is provided;
2. ensure `ext_pdu_sess` is well initialized before incrementing `payload_len` by `sizeof(*ext_pdu_sess)`.

src/gtpu/pktinfo.c

@@ -331,7 +331,11 @@ void gtp5g_push_header(struct sk_buff *skb, struct gtp5g_pktinfo *pktinfo)
         gtp1opt->NPDU = 0;
         gtp1opt->next_ehdr_type = next_ehdr_type;
         // Increment the GTP-U payload length by size of optional headers length
-        payload_len += (sizeof(*gtp1opt) + sizeof(*ext_pdu_sess));
+        payload_len += sizeof(*gtp1opt);
+    }
+    // Increment the GTP-U payload length by size of extensions length
+    if (ext_flag) {
+        payload_len += sizeof(*ext_pdu_sess);
     }
 
     /* Bits 8  7  6  5  4  3  2  1