GaussianRandomWalk for PyMC v5.6

Oh ok. I thought I had seen it somewhere given as an example of a case where the new logp inference could be used. I guess I played myself.