ODENLStepData: forward outer ODEFunction's flag to inner nlprob

[singhharsh1708]

Closes #4570.

When a user writes ODEProblem(sys, ...; jac = true, nlstep = true), the outer ODEFunction gets an analytic Jacobian but the inner teared NonlinearProblem built by generate_ODENLStepData did not — its f.jac was nothing.

This threads the jac kwarg through generate_ODENLStepData and into the inner NonlinearProblem / SCCNonlinearProblem constructor. With jac = true, MTK now symbolically differentiates the teared inner system once at build time and attaches the result to nlprob.f.jac.

Repro

using ModelingToolkit, SciMLBase
using ModelingToolkit: t_nounits as t, D_nounits as D

@parameters k₁ k₂ k₃
@variables y₁(t) y₂(t) y₃(t)
@mtkcompile rober = System([
D(y₁) ~ -k₁y₁ + k₃y₂y₃,
D(y₂) ~ k₁y₁ - k₂y₂^2 - k₃y₂y₃,
D(y₃) ~ k₂y₂^2,
], t)

prob = ODEProblem(rober,
[[y₁,y₂,y₃] .=> [1.0,0.0,0.0]; [k₁,k₂,k₃] .=> (0.04, 3e7, 1e4)],
(0.0, 1e5); jac = true, nlstep = true)

SciMLBase.has_jac(prob.f.nlstep_data.nlprob.f) # false on master, true with this PR
Verified locally
nlprob.f.jac populated when jac = true, untouched (nothing) when jac = false
The attached function is a real MTK GeneratedFunctionWrapper
Solutions still converge with the same retcode and within tolerance of the master result
Motivation
This is a prerequisite for OrdinaryDiffEq.jl-side W-matrix reuse on the nlstep path (SciML/OrdinaryDiffEq.jl#3696). Standalone performance impact is small/mixed (the inner teared problem for benchmark-scale Robertson is 1D, so FD jac is already nearly free); the real wins are downstream when paired with #3696 and on larger MTK systems where the teared inner Jacobian is non-trivial.

Checklist

• Appropriate tests were added
• Any code changes were done in a way that does not break public API
• All documentation related to code changes were updated
• The new code follows the
  contributor guidelines, in particular the SciML Style Guide and
  COLPRAC.
• Any new documentation only uses public API

Additional context

Add any other context about the problem here.

[singhharsh1708]

@ChrisRackauckas — small MTK side of the W-reuse story. Threads the existing jac kwarg through generate_ODENLStepData so the inner teared NonlinearProblem gets MTK's symbolic analytic Jacobian when the user writes ODEProblem(...; jac = true, nlstep = true). Prereq for SciML/OrdinaryDiffEq.jl#3696 closure.
Verified locally: has_jac(prob.f.nlstep_data.nlprob.f) flips false → true; same retcode on solve.

[singhharsh1708]

@ChrisRackauckas — friendly bump on this one. Re-verified: CI failures match master's pre-existing red state (same Runic failures on the same files — MTKFMIExt.jl, MTKOrdinaryDiffEqBDFExt.jl etc. — none of which are touched by this PR's 14-line jac kwarg threading). Anything else needed before merge?