Optimise habitat encoding to reduce memory footprint

[mdales]

Currently the LIFE pipeline uses Jung encoding for habitat rasters, which maps the IUCN habitat codes like 14.1 to the number 1401, etc. This is the format the Jung maps come in. However, there are less than 256 IUCN habitat classes, and yet the Jung encoding requires 16 bits to store, not 8 bits, effectively doubling our memory usage. This adds up significantly, as a Jung global map at 100m-per-pixel-at-the-equator is 150GB rather than potentially 75GB!

We should therefore re-encode the Jung maps as we transform them to our own more compact coding within the pipeline.