refactor: segment processing and mesh deduplication

[tobre1]

• Use OpenMP parallel loops to process every segment independently of thread count.
• Fix advection dissipation checks and material ID replacement.
• Rework removeDuplicateNodes() for expected linear runtime, preserving first-occurrence point data and remapping all element connectivity.
• Reuse existing index hashing and vector operations in oxidation solvers; prevent copying GPU buffer owners.
• Fix CI build configuration and make the exported VTK dependency conditional.
• Bump the version to 5.9.0, require pybind11>=3.1.0, and refresh Python stubs.