Hi @edgar,
with regard to fenics-constitutive, the procedure in fenicsx will be the same, i.e. you will still need to fill the global FE vector for the stresses (and algorithmic tangent) manually. This is also explained in more detail in the docs you referenced and the same as in the MFront approach where the idea comes from.
Regarding your concern whether this can be used for 2D and 3D
Shortly, yes, but it may depend on the problem. As far as I know @srosenbu has used this approach for 3D simulations of concrete under blast loading (explicit dynamics) and had no problems regarding memory (if this was your concern).
Fenics-constitutive
As stated on github the project is not maintained anymore for the legacy fenics. We are working on sorting stuff out towards a new version for fenicsx.
Best,
Philipp