example raster compression using Vortex

[a10y]

Just an experiment for now. Idea being, why not store raster data in Vortex.

I've run some local benchmarks extracting crops out of a local GeoTIFF.

Vortex wins by a wide margin, but it's kind of cheating because

• The raster data doesn't get compressed at all (even with Compact compressor)
• TIFF access is dominated by the DEFLATE overhead

Next steps

• Find compressor scheme that works for raster tiles, which are List<u8>
• Benchmark E2E inference pipeline VS rasterio + torch

[a10y]

I suspect we could actually squeeze out a lot of compression if we retiled into 32x32 (1024 element) vectors, and FFOR'ed each of them. Then every tile becomes a column, not a row, and each one should bitpack appropriately. This also lets us combine projection pruning for tile selection + filtering to extract a subset of a tile. Makes it very easy to turn random rectangular ranges of the image into something we can trivially push down into a scan. Plus, each 1024-element vector would probably pack relatively well.