[TRM] More automation in constitutive setting evaluation.

See merge request ogs/ogs!5001

[ci] Fixes for win jupyter ctests

See merge request ogs/ogs!5003

Could not get it to work. Fails when calling GMSH2OGS binary. Has almost
the same structure as notebook-inclined_bhe_meshing.md which runs fine.

Some improvements in LargeDeformation

See merge request ogs/ogs!5002

TH2M constitutive relations refactoring

See merge request ogs/ogs!4980

The original implementation in the assembleWithJacobian sneaked in erroneously.

Though the dmxWL/dpLR is possible, but through the code only primary variables derivatives are used, p_GR and p_cap.

This reduces number of Eigen's template instantiations and also
significantly helps with the runtime performance.

This simplifies some calls and avoids two recomputations
of the same quantity.