tsi: fix tsi_create and hijack AF_INET6

[slp]

Fix an issue where tsi_create may attempt to exit releasing sockets that hasn't been allocated.

While there, make "tsi_hijack" also hijack AF_INET6 sockets. Some applications (like ones built with recent Go libraries) attempt to bind to AF_INET6 sockets only, relying on the kernel's ability to bind on AF_INET transparently. For TSI, this means those sockets won't be hijacked and exposed to the host.

By hijacking AF_INET6 sockets, the application receives EAFNOSUPPORT when binding the socket, leading to it attempting again to bind using AF_INET.

Fixes: #99

[raphaelcoeffic]

This works for me now, thanks a lot! Starting the API as usual, which would normally bring up MPTCP and dual-stack socket. With this patch in place, I get only IPv4 TCP, as expected.

# ss -ltnp sport 3000
State              Recv-Q             Send-Q                          Local Address:Port                           Peer Address:Port             Process                                          
LISTEN             0                  4096                                  0.0.0.0:3000                                0.0.0.0:*                 users:(("VM:nuc",pid=72134,fd=566))             

[slp]

Turns our hijacking AF_INET6 and returning EAFNOSUPPORT breaks many applications, such as nginx. So we need to do the right thing and implement transparency support for AF_INET6 too.

Closing this one in favor of #101