Multi-GPU support in ESPResSo

[Jean-Noël Grad]

Dear users,

We are happy to announce the development version of ESPResSo on the 
python branch is now able to leverage multiple GPUs in LB simulations. 
This is achieved by pinning each MPI rank to a specific GPU device id. 
The exact syntax is described in the user guide:
https://espressomd.github.io/doc/lb.html#choosing-between-the-gpu-and-cpu-implementations

This is a major step forward for ESPResSo and our LB users! Some details 
still require ironing out, for example the list of CUDA devices reported 
by `system.cuda_init_handle.list_devices()` doesn't always reflect the 
true number of GPU devices when using more than 1 node in slurm, but the 
feature is mature enough to be used in production.

Best regards,
Jean-Noël Grad