Deprecate empirical point-spread function (PSF) determination

[vreuter]

It's pegged to use maximum likelihood estimation (MLE) rather than least-squares (LS), and we haven't used this in quite some time, so it's unclear how nicely it would play with all the other components and risks therefore to be buggy.

• raise NotImplementedError in the function body
• check up front in the pipeline precheck that the configuration doesn't specify to use the experimental PSF features, and if so, throw an error right away
• update the documentation accordingly