* Basic assumption: most of the data is contained in small area of volume - which is the case most of the time (1% - 2% of the volume contains 90%+ of the total brick data).
* Only the most detailed level of bricks are streamed - they are assigned to each sub level and streamed in and out with the level. The top level bricks are assigned to the persistent level so that we have something to display (and fewer discontinuities) when the sub levels are streamed out. However runtime dynamic stitching is not implemented so there might be seams in the area of streamed out sub levels - which is probably ok since the level is not needed.
* With SM4 going away, VLM sub level streaming GPU path requires SM5. All platforms having < SM5 will fall back to CPU interpolation path.
* It is recommended to enable bUseFastVoxelization=true and bUseEmbreeInstancing=true in BaseLightmass.ini with this change to speed up VLM generation especially if you have large worlds. Data management has been changed so that we can handle 4GB+ VLM data as long as you have enough video memory - however lighting build time could be not practical so use with care.
* Special notice: we have never supported the workflow of sharing sub levels between multiple worlds, or open each sub level separately to bake, and we're unlikely to support them in the future. It is possible that you happen to get some results, but they are not expected to work.
#rb daniel.wright
[CL 7979421 by Yujiang Wang in Dev-Rendering branch]
2feadce4ea