Add support for functions of multiple variables.

[GrigorySarnitsky]

As far as I understand, Optimization.jl is focused on minimizing a function f(u, p) where u had to be an AbstractArray of some reasonable type (i.e. not array of arrays).

This set up can handle functions of multiple uniform variables x1, x2, ..., we can put all these variables into one array [x1, x2, ...]. But what if the function depends on several vectors of different lengths and matricies? Although in reasonable applications it is likely that we can still put them into a single array, this is inconvinient. Flux.jl can handle functions of several array variables.

If I am wrong and Optimization.jl can handle explicitly functions of multiple variables, I would appreciate an example, couldn't find it in the docs.

[ChrisRackauckas]

Just use an array type for that, like https://github.com/jonniedie/ComponentArrays.jl

[ChrisRackauckas]

Some solvers will require flat vectors though. Handling that more nicely is something we have a solution for. SciML/DifferentialEquations.jl#881 describes some details.