UPDATE 2:
Some further testing:
- Swarm Agent logs show that it is finding the LMIV and adjusting the area and samples accordingly
- Halving the VLM detail cell size in World Settings shows an increase in brick counts/sizes in the logs
- Testing with GPU Lightmass instead of CPU does display the samples as I would expect and movable actors begin lighting more appropriately although the resulting bake is much messier
So it seems like it is finding the volume and accounting for it but the resulting volumetric lightmap is not loading or not being used in the level afterwards when CPULM is involved.