Add geometry benchmarks

[sethrj]

We know that geometry is the most expensive part of the tracking loop in real detectors. To gauge the performance of the different models through ORANGE and VecGeom (and various improvements thereto) we should model multiple simultaneous+overlapping EM showers through different full-detector calorimeter subregions, with each global subdetector grouping as a separate input.

• ATLAS EMEC
• ATLAS barrel calo
• CMS HGcal
• CMS crystal ECal

For photon problems we also want to add appropriate coupled EM-optical problems for:

• EPIC pfrich
• LHCb rich
• LZ TPC

We can use https://github.com/celeritas-project/utils/blob/main/gdml-subset/gdml-subset.cc to extract these regions and their enclosing elements independent from the rest of the geometry.

@elliottbiondo