Adjust types (and interface) such that automatons work with arbitrary python objects

[marian-lingsch]

Currently (version 9.2.0) all types and some functions like DFA.random_word indicate that the automata only work with strings for the transitions and as state identifiers.

Nonetheless, by (mis-)using this library to work with automata (NFA and DFA) for arbitrary frozen python objects and not encountering any conceptual errors (except for adjusting one line in DFA.random_word) it seems that this library is robust enough to be used with arbitrary frozen nodes and transitions for the automata.

Ideally this is explicitly reflected in the types and in the documentation.
In particular, there are likely some functions which require small adjustments DFA.random_word to work correctly with arbitrary python objects.

P.S. Great library, has been a pleasure to work with ^^

[eliotwrobson]

This is essentially a duplicate of #256. As discussed there, I don't have time to do this myself but happy to review a PR making this change. I think this could work by making the type of the accepted words a parameterized type and then using the Generic parameterization type in Python with a word being represented as an arbitrary sequence, but not 100% sure. There could be a lot of edge cases which is why I can't make time to do this myself.

I think this is also something that would be a good task to attempt with an AI agent, then we add a bunch of test cases (maybe for v10 we can include this change + refactoring of test cases when we get around to all of that).