Split register automata into structual and semantic components

[mtf90]

Currently, the RegisterAutomaton class implements DeterministicAutomaton<RALocation, ParameterizedSymbol, Transition> although its implementation doesn't satisfy the contract of a deterministic automaton. As far as I can tell, this is related to the type of input symbols that are considered.

Regarding ParameterizedSymbol, it is a non-deterministic, finite automaton. Regarding PSymbolInstance, it is a deterministic, infinite transition system (since Automatons in AutomataLib require a finite state space in order to implement Collection<S> getStates()).

One could go with an approach similar to SEVPAs which are infinite state but define additional methods for accessing its finite representation. Another approach would be to split register automata into two different views (e.g., a structural one regarding ParameterizedSymbols and a semantic one regarding PSymbolInstances where you could directly construct the semantic one from the structural one). The second approach could be interesting for planning ahead the LearnLib migration. For example, the AAARLearners also interact with concrete systems (PSymbolInstances) but return an abstract hypothesis model (ParameterizedSymbols) similar to the current RA learners. A more distinct type hierarchy could also make it easier to include different flavors of register automata (see #69).

Any thoughts?