Method dispatch_kern

dnn/src/naive/handle.h:144–151

     */
    template <class T>
    void dispatch_kern(T&& kern) {
        // this impl mainly serves to reduce binary size: we only need to
        // call ctor here, and dtor can be called from the cpp so its code
        // only needs to be generated once
        std::aligned_storage<sizeof(KernFunc), alignof(KernFunc)>::type s;
        move_kern_func_to_new_kern_and_dispatch(
                *new (&s) KernFunc(std::forward<T>(kern)));
    }

    /*!
     * \brief pass a kernel to the multi thread dispatcher associated with the