Add Markov DSL to make it a bit easier

[mattfidler]

This has some example NONMEM models that could be used to create test cases:

https://ascpt.onlinelibrary.wiley.com/doi/full/10.1002/psp4.13278

And possibly

https://pubmed.ncbi.nlm.nih.gov/9819830/

[billdenney]

This seems like it would be a simpler fit within nlmixr2lib where we have some way to add a Markov model by implied matrix operations. It would really just need a cumulative logit function to get each transition probability (maybe with a special case for binomial, single-transition models).

[mattfidler]

But it looses the ability to simulate someting closer to the input dataset and becomes simply a translation function.

[billdenney]

That makes sense. I guess my thought was that both estimation and simulation could occur by helpers. Having them directly in the DSL would be easier (though harder to maintain the DSL).