Document kwargs and fix missing uses

[gdalle]

The kwargs absstep, relstep and dir are not uniformly documented. And there might even be cases where one kwarg is forgotten, i.e. here (dir is missing). Now that DI has to use them, it would be nice if someone could explain what they do and check that they are properly passed around?

[ChrisRackauckas]

absstep and relstep are absolute and relative stepsize changes requested by the user, and dir is what direction to differentiate (i.e. what direction is "forward")

[gdalle]

is it voluntary that some functions don't use some of these kwargs, like the one I pointed to?

[oscardssmith]

I think that is probably unintentional (or at least not fully intentonal). IIUC, dir was basically hacked in by SciML to prevent calling a function out of domain so it possibly only has made it to the APIs that SciML cares about.