Particle loss across virtual surface crossings in complex regions

[yrrepy]

Bug Description

PR #3934, breaks transport with weight windows for my geometry.

The following MRE reproduces the issue
model.xml

The following errors arise:
WARNING: Could not find the cell containing particle

Steps to Reproduce

Broken

OpenMC versions ≥ https://github.com/openmc-dev/openmc/tree/03b036aa6c7231bdcf22c15de88b5e4cd3f58690

         Version | 0.16.1-dev43
     Commit Hash | c66d748caa0e55d128665c06a2c35946074ce4fe
       Date/Time | 2026-09-17 15:24:49
   MPI Processes | 1
  OpenMP Threads | 384

 Reading model XML file 'model.xml' ...
 Reading cross sections XML file...
 Minimum neutron data temperature: 1.7976931348623157e+308 K
 Maximum neutron data temperature: 0 K
 Preparing distributed cell instances...
 Writing summary.h5 file...

 ===============>     FIXED SOURCE TRANSPORT SIMULATION     <===============

 Simulating batch 1
  Primary source          particles: 1
  Secondary generation 1     tracks: 11
 WARNING:  WARNING: Could not find the cell containing particle 6
 WARNING: Could not find the cell containing particle 9
 WARNING: Could not find the cell containing particle 7
 WARNING: Could not find the cell containing particle 10
 WARNING: Could not find the cell containing particle 8
 WARNING: Could not find the cell containing particle 4
 WARNING: Could not find the cell containing particle 3
Could not find the cell containing particle 2
 WARNING: Could not find the cell containing particle 5
  Secondary generation 2     tracks: 0
 Creating state point statepoint.1.h5...

 =======================>     TIMING STATISTICS     <=======================

 Total time for initialization     = 9.4524e-02 seconds
   Reading cross sections          = 2.5890e-05 seconds
 Total time in simulation          = 2.2136e-02 seconds
   Time in transport only          = 2.0826e-02 seconds
   Time in active batches          = 2.2136e-02 seconds
   Time accumulating tallies       = 1.7500e-06 seconds
   Time writing statepoints        = 1.2722e-03 seconds
 Total time for finalization       = 2.5260e-05 seconds
 Total time elapsed                = 1.1673e-01 seconds
 Calculation Rate (active)         = 45.1749 particles/second
 Track Rate (active)               = 542.098 tracks/second

 ============================>     RESULTS     <============================

 WARNING: Could not compute uncertainties -- only one active batch simulated!
 Leakage Fraction           = 0.10000

Working

OpenMC versions < https://github.com/openmc-dev/openmc/tree/03b036aa6c7231bdcf22c15de88b5e4cd3f58690

         Version | 0.15.4-dev213
     Commit Hash | 852f92780c9c82dae7f5fbcfff91165510719f90
       Date/Time | 2026-09-17 15:25:06
   MPI Processes | 1
  OpenMP Threads | 384

 Reading model XML file 'model.xml' ...
 Reading cross sections XML file...
 Minimum neutron data temperature: 1.7976931348623157e+308 K
 Maximum neutron data temperature: 0 K
 Preparing distributed cell instances...
 Writing summary.h5 file...

 ===============>     FIXED SOURCE TRANSPORT SIMULATION     <===============

 Simulating batch 1
  Primary source          particles: 1
  Secondary generation 1     tracks: 9
  Secondary generation 2     tracks: 0
 Creating state point statepoint.1.h5...

 =======================>     TIMING STATISTICS     <=======================

 Total time for initialization     = 9.6270e-02 seconds
   Reading cross sections          = 2.3870e-05 seconds
 Total time in simulation          = 2.2712e-03 seconds
   Time in transport only          = 7.3406e-04 seconds
   Time in active batches          = 2.2712e-03 seconds
   Time accumulating tallies       = 1.3700e-06 seconds
   Time writing statepoints        = 1.4963e-03 seconds
 Total time for finalization       = 1.3437e-03 seconds
 Total time elapsed                = 1.0003e-01 seconds
 Calculation Rate (active)         = 440.289 particles/second
 Track Rate (active)               = 4402.89 tracks/second

 ============================>     RESULTS     <============================

 WARNING: Could not compute uncertainties -- only one active batch simulated!
 Leakage Fraction           = 1.00000

turning off surface checkpoints also resolves the problem (though is not a resolution of it)

    <weight_window_checkpoints>
      <collision>true</collision>
      <surface>false</surface>
    </weight_window_checkpoints>