Exclude studies with high deviance

There has been a user request to add an option to `bayes_deviance` to exclude studies with a difference in deviance between the models from the analysis. We'd need to come up with a reasonable default value but it _should_ be fairly easy to get the study names and pass them to `setup_exclude`.

Start by locating the bayes_deviance implementation and the setup_exclude call sites. Trace how study names are obtained and how exclusions are passed into the analysis, then determine and document a reasonable default deviance threshold. Done means the option excludes studies above that threshold without affecting the default analysis.

Written by the indexing model from the issue text.