Add tetrahedral grid

[arlauwer]

Description
Added the TetraMeshSpatialGrid class. This grid enables building a unique Delaunay tetrahedralization from input vertices, with a refinement option to increase quality of grid cells. This is all achieved using the open-source TetGen library. Vertices can be sampled from the medium or read from a text data file. Direct reading of tetrahedral grids is not currently supported.

Motivation
As described in Lauwers et al. (2024), adaptively constructed tetrahedral grids did not outperform existing grids in SKIRT. Nonetheless, a basic implementation of this grid can still be useful in SKIRT.

Tests
The previous iteration of this grid was extensively tested and benchmarked for accuracy. This implementation was also tested for accuracy and tested with a few edge cases, such as grids with a single tetrahedron or vertices resulting in a degenerate Delaunay tetrahedralization.

Problems

• The documentation has a minor issue where Doxygen generates class references for the private TetraMeshSpatialGrid::Tetra and TetraMeshSpatialGrid::Face classes. These classes are declared inside the TetraMeshSpatialGrid.hpp
• TetGen still displays many warnings when compiling.

[petercamps]

With my recent commits I suppressed the warnings originating from the tetgen library code for various compilers (Clang, GCC, MSVC, Intel oneAPI). Furthermore I made some cosmetic updates to the TetraMeshSpatialGrid code, including:

• always initialize local variables and data members at the point of declaration
• replace global using std::array by custom type definitions (see "using directives and declarations" in https://skirt.ugent.be/root/_dev_coding.html
• streamline the use of trailing zeroes and decimal point for floating point constants
• replace explicit copying of vector<Vec> by the built-in assignment operator
• move local variable declarations to right before where they are first used (in function storeTetrahedra)

[petercamps]

Questions remaining:

• At the end of setupSelfBefore(), mesh construction is aborted if there are fewer than 4 vertices. Rather than simply returning, I think the function should throw a FATALERROR.
• Path construction is aborted if the originating point is outside of the convex hull. This means that sources outside of the convex hull and thus also those outside of the cuboidal grid domain are ignored. This seems unacceptable (e.g. the TRUST benchmark has a single point source, which is outside of the grid domain).

[arlauwer]

After discussion with @mbaes we have decided that simply adding 8 vertices in the corners of the grid bounding box would suffice for this basic implementation. The grid will now fully cover the simulation domain and sources outside the domain are possible again. The error for not enough vertices is no longer needed as there will always be at least 8 vertices.
Additionally, the grid now removes vertices that are too close to each other. This new functionality was also tested.