fix bug: move jerry_make_api_unavailable into the end of jerry_cleanup

[jiangzidong]

In the old jerry_cleanup, it first jerry_make_api_unavailable , then call ecma_finalize ->ecma_gc_run

During gc, it may call

void
jerry_dispatch_object_free_callback (ecma_external_pointer_t freecb_p, /**< pointer to free callback handler */
                                     ecma_external_pointer_t native_p) /**< native handle, associated
                                                                        *   with freed object */
{
  jerry_make_api_unavailable ();

  ((jerry_object_free_callback_t) freecb_p) ((uintptr_t) native_p);

  jerry_make_api_available ();
} /* jerry_dispatch_object_free_callback */

which will jerry_make_api_available (); in the end

So in this way, we can't call jerry_init after jerry_cleanup, because

void
jerry_init (jerry_init_flag_t flags) /**< combination of Jerry flags */
{
  if (unlikely (jerry_api_available))
  {
    /* This function cannot be called twice unless jerry_cleanup is called. */
    JERRY_UNREACHABLE ();
  }
.....

In the patch, we move jerry_make_api_unavailable() to the end of jerry_cleanup to make sure the jerry_api_available==false after cleanup

[zherczeg]

LGTM. Nice catch

[LaszloLango]

LGTM