Win32 SD_RECEIVE has the following effects on a TCP or UDP socket:

* further calls to recv() et al. fail with ESHUTDOWN. However, Linux and Mac
  instead return 0.

* currently pending calls to recv() are not affected. However, Linux and Mac
  cause them to return 0.

* calls to WSAPoll() are unaffected. However, Linux and Mac instead signal
  both POLLHUP and POLLIN.

* for TCP connections, further data sent by the peer causes the connection to be
  reset. Mac and BSD will silently drop data, and Linux simply queues it
  forever.

In short, SHUT_RD does nothing that we want it to do on Linux or Mac (and
probably also BSD, though I am unable to test this).

Most of this impedance mismatch can be worked around in Wine, except for
WSAPoll(). Since poll() will always return POLLHUP, we can't safely ask for
POLLOUT, which we may need to do. Hence, since we aren't getting any benefit
from SHUT_RD, don't use it.

Note that this doesn't actually fix any tests, since we have some rather bizarre
logic to defer shutdown anyway (for an asynchronous socket, we don't call
shutdown(SHUT_RD) until the peer sends data; this is probably an unintended bug,
but "fixing" it ends up causing test failures, which is the main impetus for
this patch.)
Signed-off-by: Zebediah Figura <z.figura12@gmail.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>