Adjoint gradients with symmetries are only calculated for part of the domain

[smartalecH]

If an optimization problem has a prescribed symmetry condition, then gradients are only computed where the fields are physically stored (and not for any of the fields inferred by symmetry). For example, if we are designing a symmetric splitter,

then only half the gradient is computed,

whereas if we remove the simulation symmetry,

This is because we only loop over actual DFT chunks after merging #1855. Will require some thought to fix efficiently.

(cc @oskooi, @mochen4, @joelslaby)