[Nonlinear] consider passing Nonlinear.Model to solvers

[odow]

@kaarthiksundar and I talked about how it'd be good to pass the MOI.Nonlinear.Model straight to the solver (e.g., Alpine).

We kinda do at the moment, in the NLPBlockData you can access data.evaluator.model, but this might not be the most robust approach. We could add a new AbstractModelAttribute to pass the model directly, and the update JuMP https://github.com/jump-dev/JuMP.jl/blob/6a358d193d5d6e168afdaf104f5a5972add59bae/src/optimizer_interface.jl#L163-L170 to pass the model instead of the evaluator. I'm a bit worried that these are half-baked ideas though. We probably should wait until we understand the NLP expression representation at the JuMP level first.

[odow]

I think the answer is no. We should do #846 instead.