Improve autodiff docs with DI + Enzyme

[gdalle]

In #3836 we discovered that DI causes a segfault when it tries to compute a Hessian with Enzyme.
This is due to JuMP forcing splatted arguments, and Base.splat leading to type instabilities (see EnzymeAD/Enzyme.jl#1942).

I recently made some changes to DI which allow better handling of StaticArrays. If we collect the input into an SVector instead of a Vector, the segfault disappears. Should we add this trick to the tutorial, or at least a warning that Enzyme behaves differently?

Ping @wsmoses

[odow]

I'll review a PR, but I personally don't have any plans to modify the tutorial. If Enzyme doesn't work with DI, then users don't have to use it. We already include a section on how to use pure Enzyme. Adding hacks and static arrays and stuff just feels like it isn't JuMP or the user's problem.

[wsmoses]

I'll review a PR, but I personally don't have any plans to modify the tutorial. If Enzyme doesn't work with DI, then users don't have to use it. We already include a section on how to use pure Enzyme. Adding hacks and static arrays and stuff just feels like it isn't JuMP or the user's problem.

++

[gdalle]

Alrighty