First draft at host version of 1D and 2D convolution. This would not have to be fast or optimized, but just feature-parity with the GPU version.