LatticeQCD/SIMULATeQCD — reverse-engineered prompt

Build me a working version of this SIMULATeQCD project that a physicist could actually use on a GPU machine. I want the main focus to be a reliable manual build with CMake for MPI and CUDA, with support for running across more than one GPU. Please make the setup and first run experience simple, with clear docs or helper scripts so someone can compile the common executables, use the example parameter files, and run things like RHMC, quenched configuration generation, gradient flow, gauge fixing, and the plaquette example.

If anything is unclear, look up the current docs online and fill in the gaps. Please also make sure test configs that need git lfs are handled, the docs can be built, and there is an easy getting started path that shows how to choose the right GPU architecture and launch a basic run on either a local workstation or an HPC cluster.