BPF: setsockopt on an MPTCP socket: check support

[matttbe]

We have recommended to use BPF to set socket option per subflow (see #76), on TCP socket then.

But it looks like we didn't check if we can correctly do some setsockopt() on MPTCP sockets, especially the ones iterating over the different subflows. It would be good to add a test in the BPF selftests to unsure this case is covered, and verify with LOCKDEP that everything is OK there.

The reason behind that: doing a setsockopt on an MPTCP socket from BPF might be a problem if this is done in an atomic context (MPTCP will then lock the subflows, etc.)