Get rid of the arbitrary bin/lib split and instead move as much as
possible into the bin crate, which becomes the main crate again.
The types and NBT handling are moved into separate crates, so they can
be reused by nbtdump and regiondump.
Preparation for parallel processing, as well as a fix for regions
missing from later steps when the initial processing failed (rather than
using the processed data from a previous run).
Index into the biome list instead of duplicating the biome data for each
coordinate. This reduces the size of the processed data and speeds up
encoding/decoding.
This reverts commit b53d34da3d.
With the change of the mipmapper data structure, we need a conversion
step anyways, so we can keep using the Vec before mipmapping.
We want to have a sorted list in the end anyways to make metadata
generation deterministic, and we don't have to worry about deduplication
of coordinates when generating mipmap tile lists.