Fix: Const correctness in C api

[WardBrian]

As pointed out by @aseyboldt in #88, our C headers could have many more things marked as const than they currently did. This fixes that and cleans up some argument names, also per his suggestion.

[WardBrian]

Note: I didn't update the bridgestan_R header mainly because 1) it's only used by R and is redundant anywhere else and 2) I am not great at figuring out where to put the const in pointer-to-pointer types

[WardBrian]

I tested locally to confirm we cannot make param_constrain const, and with everything passing in CI I think this is good enough

[bob-carpenter]

We could get more extreme on this given that C++ lets you cast away const. The biggest problem is in our constructed model class. Despite being const, I don't believe the log_prob functions are not marked as const. What we can do in our interface is take in something that's marked const and then cast away the const modifier, which we know will be safe.

[WardBrian]

I think all of the truly const parts of the model class are marked as such. write_array requires a non-const RNG object, which is what prevents us from making it const in param_constrain

[bob-carpenter]

Nice. I hadn't realized that has been fixed. And it looks like we can't make the params arg to log_prob const because it'll be modified if it's an autodiff variable (though I'm wondering how we can declare all of our math library as const because those also get modified due to autodiff).