Remove batch information from class templates of SplineBuilder

[EmilyBourne]

In the SplineBuilder the batch information is only used in the operator()
https://github.com/CExA-project/ddc/blob/main/include/ddc/kernels/splines/spline_builder.hpp#L401-L413

A SplineBuilder is a costly object to store and initialise as it contains a matrix. It seems unnecessary to have one instance of the (same) matrix for each different batch pattern that we want to use. Further it is very verbose. We often write the following:

using SplineXBuilder = ddc::SplineBuilder<
        Kokkos::DefaultExecutionSpace,
        Kokkos::DefaultExecutionSpace::memory_space,
        BSplinesX,
        GridX,
        SplineXBoundary,
        SplineXBoundary,
        ddc::SplineSolver::LAPACK,
        GridX,
        GridVx>;
using SplineXBuilder_1d = ddc::SplineBuilder<
        Kokkos::DefaultExecutionSpace,
        Kokkos::DefaultExecutionSpace::memory_space,
        BSplinesX,
        GridX,
        SplineXBoundary,
        SplineXBoundary,
        ddc::SplineSolver::LAPACK,
        GridX>;


...
    ddc::DiscreteDomain<GridX, GridVx> meshXVx(mesh_x, mesh_vx);
    SplineXBuilder const builder_x(meshXVx);
    SplineXBuilder_1d const builder_x_poisson(mesh_x);

the code duplication and memory duplication that this involves seems unnecessary.

Wouldn't it be simpler to template operator() on the batch domain?

[tpadioleau]

I need to think of it a bit. We had a similar situation with Kokkos-fft. Some backends in Kokkos-fft require to pass arrays at the construction of the plan so that they can find the best fft algorithm. We could think of the same strategy for splines.

The plan in Kokkos-fft https://github.com/kokkos/kokkos-fft/blob/ab3221a0e03f9c8bb157ba2bff6808bc317f683d/fft/src/KokkosFFT_Plans.hpp#L67-L68