Doc unusual behavior of forward/backward methods

During their direct use (their use via FAMoS is not unusual), these methods keep trying to add parameters. i.e. if you start the forward method with the global subset model, and in the next iteration it's seen that all models with 1 greater parameter are worse, the usual forward method would terminate. However, in PEtab Select, then all models with 2 greater parameters are assessed, etc.

This is beneficial if a minimal number of parameters/processes/hypotheses are required in the model before a good fit to data is seen and thereby an improvement over the global subset model is found.

The usual forward method is the forward method used in the FAMoS method. We could add an option to choose between either the usual or unusual forward method during direct use.

Search the documentation and Python entry points for the direct forward and backward methods, then compare their behavior with the FAMoS use described in the issue. Document that direct use continues assessing models with additional parameters after the one-parameter step fails, explain why this can help, and distinguish it from the usual FAMoS method.