crash fix: theano.sparse.Dot didn't created the output var with the right number of dimensions when one input was a vector.