fix: workaround for vtkRedistributeDataSetFilter

[bd713]

This PR introduces a workaround for a regression in vtkRedistributeDataSetFilter observed in VTK versions 9.4.2 and newer.

Since VTK v9.4.2, vtkRedistributeDataSetFilter fails to correctly redistribute hybrid meshes containing tetrahedra or pyramids, resulting in silent cell loss. This issue does not seem to happen in VTK v9.3.1, but has been reproduced in both v9.4.2 and v9.5.1 (on large hybrid meshes >1M cells and at high rank counts >48), even in minimal standalones independent of GEOS.

For context, our parallel workflow relies on vtkRedistributeDataSetFilter's kd-tree capability to perform an initial, balanced distribution of any VTK mesh that is first read only by rank 0. This step is a prerequisite for parallel partitioners like ParMetis, which assume all ranks already hold a portion of the mesh.

In practice, this PR implements a "detect-and-fallback" mechanism:

• After applying vtkRedistributeDataSetFilter, the code checks whether any cells were lost.
• If cell loss is detected, it falls back to a basic scatterByBlock method, which redistributes the mesh from rank 0 using contiguous cell blocks.

As opposed to a kd-tree, this fallback approach does not rely on any geometric information, so using a partitioning tool such as ParMetis or PTScotch in post-processing is highly recommended to optimize the resulting partitions.

[MelReyCG]

We will test this one on our side.
Is there an issue on KitWare side for the kdTree issue?

[bd713]

Is there an issue on KitWare side for the kdTree issue?

Not yet, I need to coordinate with @margauxraguenel for that.

[jhuang2601]

Thanks for fixing this hidden issue.
With it, missing tet elements can be detected and recovered.