support generic device

make layers support cuda tensor (#78)

* invoke (#77)

* ASSERT_EQ

* using D = typename T0::device_type;

* fix test

* add <typename D> to layers

* simplify example

* use tensor_view as layer input

* simplify

* fix

* fix build

* concat kernel

* fix build

* io kernel

* fix include

* batch_norm kernel

* pool_traits

* pool kernel

* fix include

* typename D

* normalizer

* make it nvcc compatiable

* make it nvcc compatiable

* fix include

* make it nvcc compatiable

* stdtensor -> v0.6.1

v0.1.0

update dependency (#33)