They took the distributed game binary and ran it through a tool that produced an approximation of source code. At this point the decompiled code is clearly a mechanical transformation of the game binary, and so holds the same copyright status. This code is then cleaned up such that it compiles and functions and variables are given sensible names. This is, arguably, a creative endeavour, but it's also clearly a derivative work of the original binary and so is still controlled by copyright law.
The decompilation itself may or may not be a copyright violation - that's likely to vary depending on jurisdiction. If someone in a jurisdiction where it's legal were to write a description of the behaviour of the decompiled code and if someone else were to implement a codebase that happened to compile to the same original binary code then there's an argument that no infringement occurred (a functional description of the behaviour of a work is potentially not derivative of that work, and re-implementing the code based on that description is then not constrained by the original copyright), but that's not what happened here.
I don't like that this is the case, but let's take it to a logical conclusion - if I take an interpreted language with a bytecode compiler, I can typically decompile it to something almost 100% identical to the original code (I'd lose comments and maybe variable names, but that depends on the language). Does this mean I can take copyrighted Python code, run it through the interpreter, dump the state, decompile it, and have an independent work not subject to the original license?