snap_rounding_2 in CGAL 6.x: access violation (0xC0000005) with `Simple_cartesian<double>` but not with `Epeck`, same input

## Environment - CGAL version: 6.x - Compiler: MSVC 2022 (cl 19.4x), x64 - OS: Windows 11 ## Summary Calling `snap_rounding_2` with `Snap_rounding_traits_2<Simple_cartesian<double>>` crashes with 0xC0000005 (access violation inside the internal `Arrangement_2`) on certain near-degenerate large-coordinate datasets (coordinates ~200k–700k, pixel size 50). The exact same input completes without error on CGAL 6.x when `Snap_rounding_traits_2<Epeck>` is used instead. ## Root cause hypothesis `snap_rounding_2` appears to require exact orientation predicates for near-degenerate configurations at large coordinates. With `Simple_cartesian<double>`, floating-point rounding produces a wrong orientation sign that corrupts the internal `Arrangement_2` and leads to an access violation. `Epeck` avoids the crash because its predicates are exact. ## Workaround in use `Snap_rounding_traits_2<Epeck>` — compiles and runs correctly on CGAL 6.x, but is slower than `Simple_cartesian<double>` due to lazy exact arithmetic. ## Minimal reproducer See attached file `SnapRound_CrashRepro.cpp`. Key parameters: - 65 segments, coordinates in the range 150k–600k (millimetres) - pixel_size = 50.0 - `do_isr = false`, `int_output = false`, `kd_trees = 1` Expected on CGAL 6.x with Simple_cartesian<double>: access violation (0xC0000005). Expected on CGAL 6.x with Epeck: prints "Done." and exits normally. ## Question Is it possible to make the code work with `Simple_cartesian<double>`? Or it is inherently unstable?