fit_cluster doesn't update chisq for fixed parameters

If fit_cluster decides to skip the fit, the old value for chisq is copied over. In the existing data sets, the data has been post-processed such that the AIC etc. is NaN and chisq is computed again based on the fixed parameters (see converter.m in a5ef029 in romesh-code). This needs to be integrated into fit_cluster
