This fixes a deadlock in the Steel Series Engine installer, between the
"user_section" lock and the loader lock.

Currently user32 takes the "user_section" lock during THREAD_DETACH, i.e. while
the loader lock is held. However, it is also possible for a thread to take the
locks in the reverse order. The observed path is:
 - register_builtin()
 - LoadCursorA()
 - get_display_bpp()
 - get_display_dc()
 - CreateDCW()
 - GetDeviceCaps(DESKTOPHORZRES)
 - GetSystemMetrics(SM_CXVIRTUALSCREEN)
 - EnumDisplayMonitors()
 - update_monitor_cache()

Since we now rely on setupapi for fundamental tasks such as determining screen
size, it doesn't make much sense to delay-load it.
Signed-off-by: Zebediah Figura <z.figura12@gmail.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>