Also adds CPUOptimizedClipboard which has no extra compression
Note: Performance degrades if access is actually random (the buffering
optimizes sequential r/w)
Removing stream in favor of clipboard compression / disk
- In order to stream a schematic, it would need to be fully read 3 times
as tags are not ordered (dimensions -> block ids -> data + tiles +
entities)
- Much faster just using disk / memory as an intermediate step
