[Core] Improve reading performance StlIO

[loumalouomega]

📝 Description

This PR refactors StlIO to significantly improve reading performance by deferring node insertion and parallelising entity creation, and adds a Google Benchmark to measure the improvement (-30% time reduction aprox., depends on STL file size).

Motivation

The previous implementation created and added nodes to the ModelPart one by one inside ReadLoop, which triggered an expensive ordered-container search on every insertion. Entity creation (elements, conditions, geometries) was also done serially inside the read loop via a std::function functor passed all the way down the call stack.

How to test

# Build with benchmark support and run:
stl_io_benchmark.exe --benchmark_format=json --benchmark_out=output.json

🆕 Changelog

Performance improvements (stl_io.cpp / stl_io.h)

• Deferred node insertion – nodes are now constructed locally into a std::vector<Node::Pointer> during facet reading and added to the model part in a single bulk call after the entire solid block is parsed. This avoids repeated searches through the already-inserted node container.
• Capacity pre-allocation – before reading a solid block the stream position is saved, the file is scanned to count lines (7 lines per facet), and new_nodes is reserve()d accordingly, eliminating repeated reallocations.
• Parallel entity creation – after all nodes are collected, elements and conditions are created in parallel with IndexPartition::for_each using thread-local storage (TLS), then added to the model part in a single call. Geometry creation falls back to sequential due to an existing memory issue with parallel geometry construction.

Refactoring

• Replaced the std::function<void(ModelPart&, NodesArrayType&)> functor passed through ReadSolid → ReadFacet → ReadLoop with a plain std::vector<Node::Pointer>&, removing unnecessary indirection.
• ReadPoint now takes std::array<double, 3>& by reference instead of returning a Point value, avoiding an extra allocation and copy.
• Cleaned up stl_io.h includes (removed redundant <string>, <iostream>, <filesystem>, "includes/define.h"; added "includes/kratos_filesystem.h").
• Fixed Doxygen group from ApplicationNameApplication to KratosCore.
• Fixed typo: corresponging → corresponding.

New benchmark

• Added kratos/benchmarks/stl_io_benchmark.cpp: a Google Benchmark suite (BM_StlIO) that measures StlIO::ReadModelPart on a real STL file, with a fresh ModelPart created for each iteration.
• Added kratos/benchmarks/file.stl as the benchmark input asset.
• Updated kratos/CMakeLists.txt to build the new benchmark target.

Commits

• Refactor STL I/O headers and constructors for improved clarity and organization
• Add benchmark for STL I/O performance in stl_io_benchmark.cpp
• Refactor STL I/O to improve point reading and update method signature
• Refactor ReadModelPart to use consistent parameter naming for geometries, elements and conditions
• Refactor ReadPoint to simplify coordinate reading using range-based for loop
• Refactor STL I/O to simplify ReadSolid and ReadFacet methods, removing unnecessary functor parameters and improving node handling