Clippers don't really work. Previously we mostly didn't run into this because
the rectangle was already rejected by ddraw_surface7_Blt(), although
ddraw_surface7_BltFast() might have been affected in a couple of cases. We
should of course implement clippers, but until that happens, completely
rejecting the blit is better than introducing memory corruption.

This fixes a regression introduced by commit
92e616f3.

This actually fixes a bug. A context's current_rt field may get set to NULL if
that surface gets destroyed or unloaded. In general, the next time that
context is acquired, context_setup_target() will then set everything up again.
However, if a context_acquire() call with a NULL target ends up returning such
a context, context_setup_target() never sets the current_rt field because the
target FindContext() chooses never gets propagated to context_acquire(). This
patch ensures context_setup_target() is never called with a NULL target.