[BREAKING] feat: update to SymbolicUtils@4

[AayushSabharwal]

Requires JuliaSymbolics/SymbolicUtils.jl#795

I haven't pushed test changes since they could need multiple rounds of updates, bloating the diff and making it harder to rebase.

[SebastianM-C]

Thanks for the update, I left some questions

[SebastianM-C]

I was doing using Symbolics.SymbolicUtils to laod SymbolicUtils from the Symbolics dependency. I think that was done so that we don't have both Symbolics & SymbolicUtils as extension triggers.
Would it make sense to split the extension or to have 2 triggers given that the SU part is now more significant?

[AayushSabharwal]

Lux is the trigger, not Symbolics/SymbolicUtils. Dependencies of the parent package (Symbolics) are directly available to extensions without requiring a transitive import or additional triggers.

[SebastianM-C]

what does this do? 😅
Why don't we have x::SymbolicUtils.ShapeVecT in the function signature?

[AayushSabharwal]

promote_shape takes the operation f along with the shapes of its arguments, and returns the shape of the result. The arguments are all annotated ::ShapeT because that is the type of the shape field in BasicSymbolic. Any call to promote_shape will at best infer ::ShapeT for the shape arguments. Annotating the argument with ::ShapeVecT would mean that the compiler doesn't know at compile time whether it can dispatch to this method or not. @nospecialize tells the compiler to not compile methods for more specific types of x and ps. This helps avoid dynamic dispatch.