Improve model_comparison

1. Currently, the default for the prior odds if not specified is a probability of 1 for all models, which is improper. We should change this to 1/Number of models to be compared. The default should also be more clearly documented in the manual
2. Model_comparison should provide a warning when only ML was used. The manual should also clarify that this is only possible with Bayesian estimation
3. We might want to allow for likelihood ratio tests for nested models in case of ML