* Overhaul endian handling in ByteBuffer and FastNBT
Rather than juggling "swapped" and "unswapped" versions of integers, different library functions, #defines, etc., simply always read everything byte-by-byte.
This works regardless of host CPU endian, got optimised down to either a normal load or a byteswap on every compiler I tested - only 1 instruction on most CPU architectures.
This commit introduces a "Bytes" array type to keep endian-sensitive data seperate from host data, alongside the needed C++ template machinery for it to work seamlessly. This approach is a little bit safer as well since you get length- and type-checking for most callsites.
* Remove remaining references to old-style endianness conversion, remove functions themselves.
---------
Co-authored-by: Alexander Harkness <me@bearbin.net>
Commit is being made to allow additions of GPL3+ code previously
un-addable. With these changes, contributions back to cuberite are
possible with the backporting exemtion, as well as adding stuff in
minetest with minetest code properly being read through and implimented
to upgrade it to GPL3 from GPL2.
project still has Apache2.0 license and credits to all its contributers, but now has the freedom of GPL3+ and all the code that can be implimented and shared with it.
