Converting too/from tensors and Mandel notation would be more efficient if done in NEML

[reverendbedford]

We want a block evaluation routine that takes tensors as input, coverts to Mandel efficient, does the stress update for the whole block, and efficiently reconverts to tensors.