Standardize transform should handle edge case where variable has zero variance

In edge cases where a variable has zero variance, standardize should do the same as center rather than producing an error.

Suppose I am fitting a linear model with a large number of random covariates. I'd like to standardize them and fit a regularized model (e.g. sklearn.linear_model.Ridge). If there's some probability that a given covariate has zero variance, I'd like my code to run (and give that covariate a coefficient of zero) rather than crash.

Toy example:

import patsy
data = {"x1": [1, 2, 3, 4], "x2":[5, 5, 5, 5]}
mat = patsy.dmatrix("standardize(x1) + center(x2)", data)  # Works
mat = patsy.dmatrix("standardize(x1) + standardize(x2)", data)  # Fails! Should handle this edge case

Start by locating the implementation and tests for patsy’s standardize and center transforms, then reproduce the issue with the provided patsy.dmatrix example. Done means standardize accepts a zero-variance variable without raising an error and produces the centered result, while existing non-constant behavior remains unchanged.