reeval_internals should fix `uprev` and `uprev2` as well about delaydiffeq.jl

[devmotion]

You mean uprev and uprev2 of the ODE integrator, I guess, since the DDE integrator only needs an update of its interpolation data k, similar to
https://github.com/JuliaDiffEq/OrdinaryDiffEq.jl/blob/master/src/integrators/integrator_interface.jl#L22-L28
in OrdinaryDiffEq? These updates of t, dt, and u of the ODE integrator in reeval_internals_due_to_modification! are only to make sure that the ODE integrator (responsible for interpolation) does not interpolate on an interval that goes beyond the newly set time t of the DDE integrator with a different u. But uprev of the ODE integrator is always just updated to the final state of the solution in
https://github.com/JuliaDiffEq/DelayDiffEq.jl/blob/master/src/integrator_utils.jl#L272-L273
and if the step fails reset to the state vector of the last integration step from the solution in
https://github.com/JuliaDiffEq/DelayDiffEq.jl/blob/master/src/integrator_interface.jl#L89-L93
together with tprev.
Moreover, uprev is never copied from the solution to the ODE integrator since it is never written to - it is only used for interpolation. Finally, uprev2 of the ODE integrator is never updated in the whole algorithm so far. I assumed for interpolation uprev2 is not needed. Am I wrong? Otherwise there is no use in updating uprev2 only in reeval_internals_due_to_modification!.

[ChrisRackauckas]

Oh no, that sounds right. I was just marking this because I saw it was missing while working on the initialization. I was thinking that after a discontinuity, uprev and uprev2 would need to be handled. But doesn't make sense to use them if there was just a discontinuity, so I forgot for a second that what happens instead is the ODEIntegrator sets the booleans so that way perform_step! and all other functions know to not use the history, and it corrects itself from there. My mistake.