Implement a composition-based wrapper for SerialDenseVector

[bgoderbauer]

Description and Context

This PR migrates the current SerialDenseVector from inheritance to a true composition-based wrapper.
One major issue is that Trilinos treats a SerialDenseVector as a SerialDenseMatrix, and this is abused in a few places in our code. The new wrapper explicitly differentiates between them, so it is no longer possible to interchange them. Therefore, I had to (unfortunately) add more multiply and assemble overloads to preserve behavior. I hope we can unify or reduce them at some point. 22 different multipliers for SerialDense objects is IMO too many. Also, while working on this, I've noticed many places in the code where there are questionable decisions about the use of vectors and matrices, and their interchange. When these are fixed, the assemble methods can be reduced as well.

Some calls to base() (i.e., the Trilinos object) are still necessary until the Teuchos::SerialDenseSolver is wrapped. I will tackle this soon 🤞.

I also added the corresponding implementation for EXPECT_NEAR, which is now necessary since a vector is no longer a matrix with a second dimension of size 1.

Related Issues and Pull Requests

Follows #1209

[mayrmt]

@bgoderbauer Trilinos recommends to use Kokkos data structures for local (serial) vectors and matrices (in fact multi-dimensional arrays). I am not saying, that we should switch right away, though this might be useful information before we pour too much effort into the Teuchos::Serial stuff.

[bgoderbauer]

@bgoderbauer Trilinos recommends to use Kokkos data structures for local (serial) vectors and matrices (in fact multi-dimensional arrays). I am not saying, that we should switch right away, though this might be useful information before we pour too much effort into the Teuchos::Serial stuff.

Good to know! So, should we ultimately switch to Kokkos in the long run? With a nice abstract wrapper, it should be easy to exchange the backend.

[mayrmt]

@bgoderbauer at least, we should think about it and make an educated assessment, so that we can make a well informed decision.

[maxfirmbach]

@bgoderbauer @mayrmt I think this is a good first step. With this PR we don't inherit from Teuchos anymore, which is the right thing to do. Everything that could be done in the future is most likely worth a separate PR. I'm fine with the current state and approach.

[bgoderbauer]

I like the idea with group comments. I'll check whether it works as intended this way.

[bgoderbauer]

I think the documentation looks nice this way. What do you think?

[Copilot]

Pull request overview

This PR completes the migration of Core::LinAlg::SerialDenseVector from inheriting Teuchos::SerialDenseVector to being a composition-based wrapper, updating call sites to explicitly access the Trilinos object via base() and adding missing utility overloads to preserve existing behavior (notably in dense multiply, sparse assembly, and test assertions).

Changes:

• Introduce a composition wrapper for Core::LinAlg::SerialDenseVector and update vector APIs/usages accordingly.
• Update Trilinos solver/multiply call sites to pass vec.base() explicitly where required.
• Extend test utilities (EXPECT_NEAR) and assembly/multiply helpers to handle vectors as true vectors (not 1-column matrices).

Reviewed changes

Copilot reviewed 23 out of 23 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
unittests/common/unittest_utils/4C_unittest_utils_assertions_test.hpp	Adds EXPECT_NEAR predicate support for SerialDenseVector.
src/w1/4C_w1_input.cpp	Converts EAS vector storage to an explicit matrix wrapper when stored in history.
src/scatra_ele/4C_scatra_ele_calc_hdg.cpp	Updates SerialDenseSolver::setVectors to use trVec.base().
src/fluid/4C_fluid_DbcHDG.cpp	Replaces vector-as-matrix reshaping with vector resize.
src/fluid_ele/4C_fluid_ele_boundary_parent_calc.hpp	Updates interfaces to accept SerialDenseVector wrapper (no ::Base).
src/fluid_ele/4C_fluid_ele_boundary_parent_calc.cpp	Implements the corresponding signature updates.
src/fbi/src/4C_fbi_beam_to_fluid_mortar_manager.cpp	Adjusts kappa sizing check for vector semantics.
src/core/utils/src/numerics/4C_utils_local_newton.hpp	Updates solver RHS/unknown vectors to pass base().
src/core/utils/src/numerics/4C_utils_cubic_spline_interpolation.cpp	Updates solver vector passing to use base().
src/core/linalg/src/sparse/4C_linalg_utils_sparse_algebra_math.cpp	Updates QR solver vector passing to use base().
src/core/linalg/src/sparse/4C_linalg_sparseoperator.hpp	Adds vector overloads for sparse operator assembly (vector→column-matrix).
src/core/linalg/src/sparse/4C_linalg_sparsematrix.hpp	Adds vector overloads for sparse matrix assembly (vector→column-matrix).
src/core/linalg/src/dense/4C_linalg_utils_tensor_interpolation.cpp	Updates dense multiply calls to use SerialDenseVector::base() explicitly.
src/core/linalg/src/dense/4C_linalg_utils_densematrix_multiply.hpp	Expands multiply overload set and improves debug dimension reporting for vectors.
src/core/linalg/src/dense/4C_linalg_utils_densematrix_eigen.cpp	Replaces SerialDenseVector::Base temporaries with wrapper type.
src/core/linalg/src/dense/4C_linalg_serialdensevector.hpp	Introduces the new composition-based SerialDenseVector wrapper API.
src/core/linalg/src/dense/4C_linalg_serialdensevector.cpp	Implements wrapper methods and keeps free norm2/update utilities working.
src/core/linalg/src/dense/4C_linalg_fixedsizematrix.hpp	Updates documentation reference from SerialDenseVector::Base to wrapper.
src/core/fem/tests/geometry/4C_fem_geometry_element_volume_test.cpp	Updates test code from reshape(...,1) to vector resize.
src/core/fem/src/general/utils/4C_fem_general_utils_polynomial.cpp	Updates solver vector passing and vector sizing to match new wrapper semantics.
src/core/fem/src/discretization/4C_fem_discretization_hdg.cpp	Replaces vector-as-matrix reshaping with resize.
src/beaminteraction/src/crosslinking/4C_beaminteraction_crosslinking_submodel_evaluator.cpp	Replaces vector-as-matrix reshaping with resize.
src/art_net/4C_art_net_art_junction.cpp	Updates solver vector passing to use base().

[mayrmt]

@bgoderbauer Looks good overall! I left some comments for possible further improvement within this PR.