[Core] Optimize ComputeTriangleLineIntersection and PointInTriangle

[loumalouomega]

📝 Description

This PR optimizes the hot-path intersection utilities identified via VTune profiling during meshing operations. The profiling showed ComputeTriangleLineIntersection and boost::numeric::ublas::scalar_minus consuming a significative amount of time in a typical meshing run.

83% time reduction in ComputeTriangleLineIntersection

(BTW, Ublas should be avoided, many performance issues come from it...)

🆕 Changelog

Changes

1. Direct Scalar Operations in ComputeTriangleLineIntersection

Replaced boost ublas vector operations with explicit scalar arithmetic:

• Eliminated 5 array_1d<double,3> temporary allocations
• Replaced MathUtils::CrossProduct, inner_prod, and vector subtraction with direct component-wise operations
• Uses squared norm check to avoid sqrt() call for degenerate triangle test

2. Inlined Barycentric Coordinate Check

Instead of calling PointInTriangle separately, the barycentric coordinate test is now inlined:

• Reuses already-computed edge vectors u and v
• Eliminates redundant vector subtractions that were duplicated in PointInTriangle

3. Optimized PointInTriangle

Replaced boost ublas operations with direct scalar arithmetic:

• Explicit component-wise vector subtraction
• Direct dot product computation without inner_prod overhead

Performance Impact

Optimization	Expected Savings
Direct scalar ops	Eliminates boost ublas temporary allocation and operator overhead
Inlined barycentric	Avoids redundant edge vector computation

Commits

• Optimize ComputeTriangleLineIntersection and PointInTriangle
• Add benchmarks for triangle-line intersection and point-in-triangle utilities