Refine MSC/along-step implementation

[sethrj]

• [x] Vecgeom crashes on GPU and produces different results from ORANGE (on host) in the stepper test
• [x] is_displaced, geo_limited, action_result are not orthogonal choices
• [x] range is recalculated a couple of different time; and there's a min which is necessary only if the range recalculation changes or running some kind of test 454, 429, 398: fixed by #467
• [x] (after #457) move urban MSC out of core physics, after refactoring energy loss and data in #253
• [x] Move fixed-energy cutoff (#454) for charged particles into CutoffParams and apply to produced secondaries, and possibly after slowing down tracks?
• [ ] Skip zero-step check if we're applying discrete interactions (https://github.com/celeritas-project/celeritas/pull/457#discussion_r924802148)
• [ ] Revisit whether to use an average speed and physics/geometry step length (https://github.com/celeritas-project/celeritas/pull/457#discussion_r924827863)
• [x] Understand/address failing assertions:
  • [x] /home/alund/celeritas_project/celeritas/src/celeritas/global/alongstep/AlongStep.hh:120: celeritas: internal assertion failed: mfp > 0 (positron with a boundary action and zero mfp) (#642)
  • [x] SBEnergyDistribution.hh:168: celeritas: internal assertion failed: xs >= 0 && xs <= 1 / inv_max_xs_ (on NVIDIA GeForce RTX 3090, sampled gamma energies in brems from incident positrons are below the production cut)
• [x] Generalize geo↔︎phys length conversion
• [ ] Change step limiter to be a separate pre-step kernel
• [ ] Use physics helper functions for range/eloss instead of re-implementing in MSC helper
• [ ] Allocate temporary "MSC" step view with range cache, msc interaction result, geometry step length, ...
• [ ] Factor out the true/geom path length conversion

[sethrj]

See #615