Looping over nested fields

This is related to issue: #337

Let's say we have the ComponentArray:

n_cells = 100

u = ComponenArray(mass_fractions = (methanol = zeros(n_cells), water = zeros(n_cells))

And we wanted to loop over mass fractions to apply species_diffusion, advection, a chemical reaction, etc.

for cell_id in 1:n_cells
    for species_name in u.mass_fractions
        u.mass_fractions[species_name][cell_id] += 1.0
    end
end

Well right now, this pattern is terrible for performance and also doesn't work with Polyester's @batch.

I've also tried:

for cell_id in 1:n_cells
    map(keys(u.mass_fractions) do species_name
        u.mass_fractions[species_name][cell_id] += 1.0
    end
end

but this also performs terribly and also performs terribly with polyester.

Pure NamedTuples perform really well when not using @batch. For example:

u = (mass_fractions = (methanol = zeros(n_cells), water = zeros(n_cells))

for cell_id in 1:n_cells
    map(keys(u.mass_fractions) do species_name
        u.mass_fractions[species_name][cell_id] += 1.0
    end
end

doesn't allocate at all and performs amazingly, but if you add @batch to the loop, it allocates and performance goes down the drain. However, NamedTuples aren't ideal as non-vector quantities stored inside a single field cannot be mutated.

If you want to see the performance differences see here:
https://gist.github.com/Jolt8/3b7cf79326f9ffbee6626a0d39e6e428

If there was a dedicated method for looping over nested fields in a ComponentArray that also worked with Polyester, that would be absolutely amazing!

Start by reading related issue #337 and the linked performance gist, then investigate how ComponentArray nested fields are currently iterated alongside Polyester's @batch. Done should be a dedicated nested-field iteration approach that avoids the reported allocations and works with @batch.