add specification for lrzip

[armijnhemel]

This PR adds a specification for the lrzip format. It mostly works, except that if there is a trailing MD5 sum it isn't easy to automatically recognize it because it is necessary to first find out where the data ends and propagate that (and I don't know how to do that cleanly with instances).

Also, this one will not work with very big files, as I was too lazy to implement something like b40le, b48le and b56le, so there is still room for improvement.

[KOLANICH]

I guess

switch-on: byte_width
cases:
  1: u1
  2: u2
  3: b24le
  4: u4
  8: u8

can be an own type.