Refactor replica types to make it easier to add new ones

Current replica types are integrated in the code in an ad-hoc manner. Consequently, it is not easy to modify the behaviour of an existing replica type or to add a new one.

I'm working on separating compute from storage (with Ilan and others), and for the purpose of this project we would need to add a new replica type (or maybe to alter the behaviour of existing ones). This is why I would like to propose some iterative refactoring steps to make it possible to do so with less boilerplate.

First PR is around making it easier to work with replica counts, i.e., the number of replicas per replica type. It avoids tracking individually number of replicas of a given type via individual variables.