Adding a homogenization scheme

MeanFieldHom.Schemes ships Voigt/Reuss, dilute (direct and dual), Mori-Tanaka, Maxwell, Ponte-Castañeda–Willis, self-consistent (symmetric and asymmetric) and differential, each in an elastic and an ageing-viscoelastic flavour. A new scheme slots in beside them:

Create src/Schemes/<scheme_name>.jl and include it from src/Schemes/Schemes.jl.
Declare the scheme type (struct MyScheme <: HomogenizationScheme) in src/Schemes/scheme_types.jl, and register its symbol alias in SCHEME_ALIAS (src/Schemes/homogenize.jl) if you want homogenize(rve, :myscheme, :C) to work.
Implement _evaluate(rve, ::MyScheme, ::Val{p}; kw...). Provide one method per tensor order — dispatch on the order of matrix_property(rve, p) — so the scheme serves elasticity and transport from one implementation, as _mt_dispatch does in mori_tanaka.jl.
Go through contribution_helpers.jl. Never call stiffness_contribution and friends directly on a phase geometry: the _phase_* helpers apply the amount, honour the symmetrize setting, hand the kernel the correctly pre-projected reference medium, and branch on is_homogeneous_inclusion. Two invariants stated in that file's header must hold in your kernel too — all helpers of a given evaluation must share the same projected P₀, and orientation averaging must never be folded into a tensor product.
Prefer the bundled seams (_phase_dilute_and_contribution, _phase_dilute_and_stress_average, _phase_compliance_and_contribution) when you need two objects per phase: they share the single expensive solve.
Consider what your kernel requires of an inclusion. Needing only the contribution tensors keeps the scheme usable with inclusions entered through gate C of the inclusion contract; needing the concentration tensor restricts it.
Export through src/Schemes/Schemes.jl and re-export from src/MeanFieldHom.jl.
Add a unit test under test/Schemes/ and a section in docs/src/manual/schemes.md.