Fix for bug #12670 (https://gna.org/bugs/index.php?12670).
This bug was reported by Michael Funk (mifu att nmr dot mpibpc dot mpg dot de).
The problem was that the relaxation curve-fitting return_error() method was returning the global
cdp.sigma_I error structure - which does not exist for all error types - when it should have been
returning the spin specific error structure.
