support of non-adaptive algorithms

[acroy]

Mainly to facilitate the discussion of the API in #7, I opened this issue on non-adaptive algorithms. The question was if such algorithms (like RK4) should be part of ODE.jl and how their interface should be designed to be useful.

[acroy]

@pao:

There are no intermediate times. You're only integrating one time step into the future. Your system will have changed due to the influence of the control system for the next time step.

I am not sure I understand "your system will have changed" in this context, but would it work for you if there was an option to do the full step with ode45 and return the estimated error instead of refining the time step? This would also involve a fixed number of evaluations of the RHS.

[ChrisRackauckas]

I believe this was fully solved by #68.