Use the SciMLStructures Tunable interface

[ChrisRackauckas]

This uses the SciMLStructures Tunable interface https://github.com/SciML/SciMLStructures.jl in order to allow more generalized definitions of p.

• Ensure Lux.jl is well supported (componentarrays extension in SciMLStructures
• Add a test for a custom SciMLStructure

[ArnoStrouwen]

You need to actually pass repack?

SciMLSensitivity.jl/src/concrete_solve.jl

Line 219 in dcf961a

default_sensealg = automatic_sensealg_choice(prob, u0, p, verbose)

[ChrisRackauckas]

Yeah this isn't quite done. We should do a branch of, if it's a mutable structure then we just use replace!(::AbstractPortion, p::T1, new_values), and only repack if it's immutable.

[ArnoStrouwen]

Something isn't completely clicking with me, how this change can be so simple.
Where is the logic living that when in the adjoint equations df/dp and dg/dp occurs, we should only take the tunable parameters?

[ChrisRackauckas]

I should've marked WIP. Plan to finish it during today.

[xtalax]

Can we expect this to be released by Monday?

[ChrisRackauckas]

yes