Do the target-waiting within do_initial_child_stuff on Windows.

This is a preparatory patch that achieves two goals:

  . Makes the initial event handling more similar to GDB's;
  . Opens the door for implementing post-inititial-handling
    operations.

At the moment, this is only done on Windows, where the
post-initial-handling is going to be needed (in the context of
Windows 2012). And because we're close to creating the gdb 7.7
branch, making that change for all platforms is a little more
risk that we'd like. So the change is currently implemented
on Windows.

gdb/gdbserver/ChangeLog:

        * target.c (mywait): Set OURSTATUS->KIND to TARGET_WAITKIND_STOPPED
        if equal to TARGET_WAITKIND_LOADED.
        * win32-low.c (cached_status): New static global.
        (win32_wait): Add declaration.
        (do_initial_child_stuff): Flush all initial pending debug events
        up to the initial breakpoint.
        (win32_wait): If CACHED_STATUS was set, return that instead
        of doing a real wait.  Remove the code resuming the execution
        of the inferior after receiving a TARGET_WAITKIND_LOADED event
        during the initial phase.  Also remove the code changing
        OURSTATUS->KIND from TARGET_WAITKIND_LOADED to
        TARGET_WAITKIND_STOPPED.

gdb/gdbserver/target.c

@@ -82,6 +82,11 @@ mywait (ptid_t ptid, struct target_waitstatus *ourstatus, int options,
 
   ret = (*the_target->wait) (ptid, ourstatus, options);
 
+  /* We don't expose _LOADED events to gdbserver core.  See the
+     `dlls_changed' global.  */
+  if (ourstatus->kind == TARGET_WAITKIND_LOADED)
+    ourstatus->kind = TARGET_WAITKIND_STOPPED;
+
   /* If GDB is connected through TCP/serial, then GDBserver will most
      probably be running on its own terminal/console, so it's nice to
      print there why is GDBserver exiting.  If however, GDB is