fix: propagate tofloat, use_union to better_varmap_to_vars

[vyudu]

Close #3446

[ChrisRackauckas]

Looks like this is giving some test failures?

[vyudu]

The test failures seem unrelated

[vyudu]

I can't reproduce these failures locally

[AayushSabharwal]

We should only do this propagation for JumpProblems. A union-typed u0 is not valid for ODEs iirc which looks like the cause of this error.

[vyudu]

Shouldn't it just error if someone passes use_union = true for an ODEProblem then? It doesn't make sense for someone to pass in use_union = true and it just gets ignored right

[vyudu]

Or should it propagate for the p but not the u0?

[AayushSabharwal]

Yeah it needs to propagate for the p

[vyudu]

Is this good to merge?

[isaacsas]

Is use_union really relevant or needed anymore? Before it was needed to avoid parameters being converted to integers but that is now handled by making them symbolic integers. All jump systems really need is to preserve integer initial conditions as integers, not as a union type, without requiring users to specify the symbolic variables as integers originally.

[ChrisRackauckas]

Yes we should probably error for any unions. We created the split parameters to handle that, so if we end up with unions anywhere that's a sign something went wrong.

[AayushSabharwal]

So use_union is removed as a kwarg and defaults to false? Sounds like a nice idea and simplifies some things.

[vyudu]

I can add that.

[vyudu]

Actually we'd still need use_union to handle cases like #1561 right?

[AayushSabharwal]

varmap_to_vars should remain untouched. We don't use it anyway, it's just for backward compat.

[vyudu]

What should become of the final test here that tests if p has a union eltype?

ModelingToolkit.jl/test/split_parameters.jl

Lines 107 to 128 in 0ad773d

	vars = @variables y(t)=1 dy(t)=0 ddy(t)=0
	pars = @parameters a=1.0 b=2.0 c=3
	eqs = [D(y) ~ dy * a
	D(dy) ~ ddy * b
	ddy ~ sin(t) * c]
	@named model = ODESystem(eqs, t, vars, pars)
	sys = structural_simplify(model; split = false)
	tspan = (0.0, t_end)
	prob = ODEProblem(sys, [], tspan, []; build_initializeprob = false)
	@test prob.p isa Vector{Float64}
	sol = solve(prob, ImplicitEuler());
	@test sol.retcode == ReturnCode.Success
	# ------------------------ Mixed Type Conserved
	prob = ODEProblem(
	sys, [], tspan, []; tofloat = false, use_union = true, build_initializeprob = false)
	@test prob.p isa Vector{Union{Float64, Int64}}

Removing use_union would break it

[AayushSabharwal]

Couple minor things, and needs a format