Update cluster_theory_pred to match CLPFirecrown and add a tutorial

[eduardojsbarroso]

clpipe/cluster_theory_pred.py was out of date with clp_firecrown.py and the redmapper example. It used a hard-coded cosmology (sampled omega_c and sigma_8 were silently dropped), different defaults and purity model, and never called the recipe setup. On the redmapper data the counts came out 5 to 10% low.

The module now builds the same likelihood as the likelihood_file.py written by CLPFirecrown and evaluates it with Firecrown at one point. The defaults come from CLPFirecrown.config_options and the cosmology from the fiducial cosmology file plus the cosmological_parameters overrides. Parameters take the values written in priors_file.ini (starting value when sampled), or the ones passed in params, for example a chain best fit. Chain column names like cosmological_parameters--omega_c also work.

from clpipe.cluster_theory_pred import compute_cluster_predictions

pred = compute_cluster_predictions(config_file, sacc_file, fiducial_cosmology, params=None)
pred.counts.theory, pred.counts.data, pred.counts.errors, pred.shear.theory, pred.chi2

build_cluster_recipes_from_config is kept but now needs fiducial_cosmology=.

The new tests compare the predictions with the likelihood and values files generated by CLPFirecrown on the mock SACC. I also checked on CC with firecrown_developer_clp against the likelihood_file.py of the redmapper baseline run, and got the same theory vector and chi2 (246.2).

The tutorial is in tutorials/theory_prediction_tutorial. It compares the redmapper baseline data with the predictions at the starting point and at the chain best fit, and shows a quick hmf comparison. It needs the example data from the portal (see README).

Two things to keep in mind. There will be a small conflict with #39 in the docstring of cluster_theory_pred.py, keep this version. Also use_grid: false is not covered by the comparison test yet, because clp_firecrown.py on main does not import ExactBinnedClusterRecipe (fixed in #39).