Linearization should not merge all enumeration types

Currently linearization transformations (product and merge variants) merge all enumerations together to a single enumeration. This includes the enumerations created during the transformation, for the locations of automata.

This is not desired, as it makes it less clear what values the variables with enumeration types can really have. Also, it makes the domains of the variables larger, which can lead to performance issues for certain operations on the models (e.g. synthesis or formal verification).

We should not apply the MergeEnums CIF to CIF transformation as part of linearization. We may have to ensure that all enumerations get proper unique names in the result, etc.