Fix sefault in pgpool child process in certain case.

Fix sefault in pgpool child process in certain case.

When all backend go down, pgpool refuses to accept connection from
clients and try to send a message in
validate_backend_connectivity(). For this purpose it connects to the
client and try to read the startup packet. This is done in a PG_TRY
block. The startup packet is read in memory pointed to by "sp", which
is declared as "StartupPacket *volatile". This is fine. But it was
forgotten to initialize the value with NULL. As a result, if reading
startup packet fails, sp would be garbage a pointer and segfaults later
in pool_free_startup_packet(sp).

Fix is, initialize "sp" with NULL.

I have found this accidentally in following way:

1) shutdown all backends.

2) connect to pgpool with invalid client. I have used pcp_attach_node
with pgpool's port number, not pcp's.