feat: improve trace setup reuse and benchmarking support

[tobre1]

• Add explicit commitGeometry() support for CPU disk and triangle tracing, with cached Embree scene/boundary setup and invalidation when geometry, boundary conditions, or source direction change.
• Improve tracing performance by removing virtual geometry dispatch, templating the trace kernel on concrete geometry types, using stack-backed disk hit storage, and parallelizing output reductions.
• Add configurable max backface-hit handling and expose new trace statistics for backface hits and average disk hits.
• Fix free-flight scattering by sampling and updating a physical free-flight distance across ray segments.
• Add GPU benchmarking support behind VIENNARAY_BENCHMARK, including trace-count tracking APIs and CMake/OptiX rebuild plumbing.
• Split source generation into separate getOrigin() and getDirection() APIs, and update tests/examples accordingly.