Improve HPC documentation beyond SLURM

[ncclementi]

xref: rapidsai/docs#572

The HPC docs would benefit of a general update beyond SLURM which will be tackled by #54
and probably include a Troubleshooting section.

Common problem: There is a challenge with installing CUDA-enabled conda packages of PyTorch (and some other libraries) on CPU-only nodes, often HPC login nodes. This does not affect RAPIDS libraries but it comes up often in RAPIDS+PyTorch envs. Therefore documenting this to have an asset to point people to seems reasonable.

cc: @bdice for visibility