Make AtlasG4 and FullG4 simulation identical

Address difference seen in ATLASSIM-4656 between the AtlasG4 simulation and the FullG4 simulation inside ISF by changing the definition of the G4Material Vacuum that is created for a Fatras only dummy detector. In master, this definition was picked up also for standard G4 jobs in ISF, causing a different Vacuum to be used.

FYI @abasalae, nightly regression tests for FullG4 and all other ISF simulations running G4 will likely fail because a different Vacuum material is being used!

Please merge only one nightly after !35238 (merged), so we see the ART changes of both independently!