RODE analytic from full solution not set up for dense errors

[rmsrosa]

Recently, the computation of an analytic solution to a RODE problem got the ability to be computed from the full solution #233 .

This works well when computing the timeseries_errors (at rode_solutions.jl#L116).

However, if dense_errors is set to true, then f.analytic_full is not checked and the solver errors when attempting to interpolate the analytic solution using the standard API for sol.prob.f_analytic (at rode_solutions.jl#L144).

My suggestion is to define f_analytic!(sol, times=sol.t) (or f_analytic!(sol, times=sol.W.t, noises = sol.W.W) so we can use sol.prob.f.analytic(sol), when computing the time series error, and call sol.prob.f.analytic(sol, densetimes) or sol.prof.f.analytic(sol, densetimes, sol.W(densetimes)), when computing dense_errors. I can work on that when I get the time.

[ChrisRackauckas]

That sounds reasonable.