`DirectNavigator` can fail in rare cases with `CylinderSurfaces`

[andiwand]

After diving into physmon changes of #4694 and #4887 I believe we have a rare edge case with the DirectNavigator and CylinderSurfaces if the trajectory deviates from the initial one which estimated the surface sequence.

The problem seem to come from the fact that we disambiguate intersection solutions on the fly due to the usual CylinderSurfaces problem and the fact that the surface order can change with infinite bounds. In some cases the correct cylinder intersection will then be behind the current position and be rejected and the second solution is accepted. In such a case the propagation will practically aim for the other side of the detector and the fit will not be successful.

[benjaminhuth]

This also affects box detectors built out of Rectangle surfaces.

I think the solution would be to make the DirectNavigator aware of volumes, either by querying the other boundary surfaces from the geometry in case the next surface in the stream is a boundary surface, or by supplying potential additional boundary surfaces as a map by the user.

This would also allow to hack reliable navigation without a geometry which could be nice for tests or quick experiments.

[github-actions]

This issue/PR has been automatically marked as stale because it has not had recent activity. The stale label will be removed if any interaction occurs.