Fix parameter dependencies across model hierarchy

[SebastianM-C]

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

Parameter dependencies can get broken if parameter_dependencies(sys::AbstractSystem) returns Vector{Any}, as process_parameter_dependencies(pdeps, ps) checks the element type

ModelingToolkit.jl/src/systems/abstractsystem.jl

Lines 2951 to 2953 in 874985c

	if !(eltype(pdeps) <: Equation)
	error("Parameter dependencies must be a `Dict`, `Vector{Pair}` or `Vector{Equation}`")
	end

This PR adds a type assert in equation namespaceing and makes parameter_dependencies infer better.

julia> @code_warntype ModelingToolkit.parameter_dependencies(sys1)
MethodInstance for ModelingToolkit.parameter_dependencies(::ODESystem)
  from parameter_dependencies(sys::ModelingToolkit.AbstractSystem) @ ModelingToolkit ~/.julia/dev/ModelingToolkit/src/systems/abstractsystem.jl:1342
Arguments
  #self#::Core.Const(ModelingToolkit.parameter_dependencies)
  sys::ODESystem
Locals
  #1409::ModelingToolkit.var"#1409#1411"
  namespaced_deps::Vector{Equation}
  systems::Vector{ODESystem}
  pdeps::Vector{Equation}
Body::Vector{Equation}
1 ─       Core.NewvarNode(:(#1409))
│         Core.NewvarNode(:(namespaced_deps))
│         Core.NewvarNode(:(systems))
│         Core.NewvarNode(:(pdeps))
│   %5  = ModelingToolkit.:!::Core.Const(!)
│   %6  = ModelingToolkit.has_parameter_dependencies(sys)::Core.Const(true)
│   %7  = (%5)(%6)::Core.Const(false)
└──       goto #3 if not %7
2 ─       Core.Const(:(Base.getindex(ModelingToolkit.Equation)))
└──       Core.Const(:(return %9))
3 ┄       (pdeps = ModelingToolkit.get_parameter_dependencies(sys))
│         (systems = ModelingToolkit.get_systems(sys))
│   %13 = ModelingToolkit.:(var"#1409#1411")::Core.Const(ModelingToolkit.var"#1409#1411")
│         (#1409 = %new(%13))
│   %15 = #1409::Core.Const(ModelingToolkit.var"#1409#1411"())
│   %16 = (:init,)::Core.Const((:init,))
│   %17 = Core.apply_type(Core.NamedTuple, %16)::Core.Const(NamedTuple{(:init,)})
│   %18 = Base.getindex(ModelingToolkit.Equation)::Vector{Equation}
│   %19 = Core.tuple(%18)::Tuple{Vector{Equation}}
│   %20 = (%17)(%19)::@NamedTuple{init::Vector{Equation}}
│   %21 = systems::Vector{ODESystem}
│         (namespaced_deps = Core.kwcall(%20, ModelingToolkit.mapreduce, %15, ModelingToolkit.vcat, %21))
│   %23 = namespaced_deps::Vector{Equation}
│   %24 = pdeps::Vector{Equation}
│   %25 = ModelingToolkit.vcat(%23, %24)::Vector{Equation}
└──       return %25

One case that shows this issue is having parameter dependencies that reference a subsystem (and this didn't have a test anyway, so I added that).

Note that while #2966 also made changes in this area, the test case that I am adding was not fixed by that. I had an initial version of this PR which would split off the empty case, but after the recursion, but splitting that off should no longer be needed as type inference on mapping namespace_equation is better and we should always have Vector{Equation}.

[AayushSabharwal]

This is failing CI though

EDIT: Wait, master errors this way too.

[SebastianM-C]

@ChrisRackauckas I rebased the PR

[AayushSabharwal]

Format?

[SebastianM-C]

I saw that you're doing it in 2a298d0, so I didn't do it to avoid conflicts 😅

[AayushSabharwal]

Conflicts are a problem for whoever merges last :P Also, in this specific case that commit would just end up being dropped