MACE-LAMMPS meta issue

[wcwitt]

I need a place to organize/triage some LAMMPS-related issues.

• better tests for different architectures!
• need to document multi-GPU Request for Documentation Update - Multi-GPU and Single Precision Support in MACE-LAMMPS #274
• this PR needs merging LAMMPS multiple MPI instances over limited GPUs (multiple repllica simulations) lammps#1 (and related LAMMPS on multiple GPUs #195 (comment))
• need to tidy/improve single-precision support for Kokkos Request for Documentation Update - Multi-GPU and Single Precision Support in MACE-LAMMPS #274
• need per-atom energies in Kokkos mode LAMMPS-MACE compute error #193
• a request for per-atom virials Atomic virial calculation in lammps #272
• investigate Kokkos::parallel_scan to tidy the Kokkos code (suggested by Anders Johansson)
• refactor the MPI splitting using something like MPI_COMM_WORLD_rank % n_gpus_per_node (suggested by @bernstei)
• request for plug-in mode (need to think about this one) MACE LAMMPS in plugin mode #271
• problem compiling with MPI turned off (reported by Philip Pracht)

[ilyes319]

Excellent idea, thank you!