Make the Magicavoxel parser load voxels lazily

[DouglasDwyer]

In my engine, I use gvox as follows. I identify chunks of the world (256³ regions) to be loaded, and forward them to background threads. These background threads instantiate a gvox context and adapters, and then do a parser-driven blit for that chunk.

The Magicavoxel adapter, upon instantiation, reads and loads the entire Magicavoxel file into memory as flat arrays. This is very wasteful and leads to computer lag for my use case (the system allocator is heavily burdened), since a bunch of extra models are loaded that fall outside of the 256³ region about which I care.

To fix this, this PR adds lazy loading of Magicavoxel model chunks. The voxels for a model from a Magicavoxel file are not decompressed until that model is read for the first time. This improves performance and eliminates lag spikes for me.

[GabeRundlett]

Ignore the checks. This looks good