* optimize content-dnn cuda kernel

- add var_conv_2d cuda kernel

- add var_conv_2d cuda kernel unit test

- temporarily set to two input mode, remove input(ROW) and input(COLUMN)