Run pf divert-to and divert-reply regression test.  The framework
runs both a client and a server process.  One process is started
on the local and the other one on the remote machine.  The kernel's
pf of the remote machine gets tested.  The remote machine's pf.conf
must contain an anchor named "regress" where the test places its
divert rules automatically.  All tests are done with IPv4 and IPv6.

The protocols TCP, UDP, Raw IP, ICMP get tested.  TCP uses a listen
and a connected stream socket, the others use bound and possibly
connected datagram sockets.  Over the TCP connection messages are
sent on both directions.  The datagram sockets deal with single
packets.  The remote machine can be reached over a non existing
address which is diverted to the stack.

The args-...-to tests install an incoming divert-to pf rule on the
remote machine and run a server there.  The server binds to the
localhost address, for TCP it also listens and accepts.  The client
is started on the local machine and connects to the non existing
address of the remote machine.  For TCP the bidirectional connection,
for the others a singe packet from the client has to reach the
server.

The args-...-reply tests install an outgoing divert-reply pf rule
on the remote machine and start a client there.  The client binds
with bindany to the non existing address and connects to the local
machine's address.  The server is run on the local machine, for TCP
it also listens and accepts.  For TCP the bidirectional connection,
for the others a singe packet from the client has to reach the
server.

The args-...-reply-to tests use the same setup as the args-...-reply
tests.  But addtitionally to the packet from the client to the
server, the server sends a packet back which has to be received by
the client.  To figure out the client's address and port, the server
receives with recvfrom and sends back with sendto.  The args-icmp-reply-to
test does not use a server as the kernel of the local machine
automatically reflects the ICMP echo request packet with an reply.

The args-icmp-reply-reuse test is similar to the args-icmp-reply-to
test, but it sends two ICMP echo requests and expects two ICMP echo
replies.  All four packets use the same socket.  The second echo
has a different ID, so it cannot use the same pf state.  Check that
the second reply reaches the client.  This can only work, if pf
creates a second outgoing state although all packet use one socket.

The reuse tests run pairs of corresponding args-...-to and
args-...-reply and args-...-reply-to tests consecutively to check
that the pf states to not interfere.  The first run flushes the
state, the second must get rid of the state automatically.  For TCP
the connection in TIME_WAIT is dropped to remove the state.