OrdinaryDiffEqVerner 1.12.0 breaks installation

[duetosymmetry]

Precompilation works with OrdinaryDiffEqVerner 1.11.0, but fails with 1.12.0

[duetosymmetry]

Sorry I don't have any more information about the underlying reason this now fails. I'm not sure if changing the compat to limit the version of OrdinaryDiffEqVerner is correct, or if there is a better fix that would allow using the latest version.

[duetosymmetry]

So far I tracked this down to the following. In SciML/OrdinaryDiffEq.jl@840c669, @ChrisRackauckas switched lazy from being a Bool to being a Val{true} or Val{false}. Now in integrator_interface.jl:SciMLBase.reeval_internals_due_to_modification!, we have the line:

            ode_addsteps!(integrator, integrator.f, true, false, !_unwrap_val(alg.lazy))

which for some reason is evaluating as !(::Val{true}), which is not defined. Maybe alg.lazy ended double-wrapped as Val{Val{true}}? Not sure how this could have happened.

EDIT: Sorry, it's not double-wrapped. Something in our env was trying to install OrdinaryDiffEqVerner 1.12.0 with an older version of OrdinaryDiffEqBase, which wasn't doing the unwrapping.

[duetosymmetry]

Something in our deps is forcing OrdinaryDiffEqCore to 3.1.0 (I think because some dep of newer OrdinaryDiffEqCore wants a newer FastDifferentiation). Meanwhile OrdinaryDiffEqVerner@1.12.0 does not have a compat statement saying it needs OrdinaryDiffEqCore>=3.5. So Julia thinks it satisfy all the deps with OrdinaryDiffEqVerner@1.12.0 and OrdinaryDiffEqCore@3.1.0 (which doesn't unwrap alg.lazy), hence this bug.

If we can figure out how to use newer FastDifferentiation, we can avoid this problem.

[duetosymmetry]

Oops, sorry, I left a comment here that should have actually gone on #110 . See my comment there which un-breaks the dependency issue with OrdinaryDiffEqVerner