* add nearest_interp_cuda kernel, test=develop

* add concat op and elementwise_add op

* remove eigen dependency from nearest_interp cuda kernel, test=develop

* free cuda pointers, test=develop