Add HWLOC_KEEP_NVIDIA_GPU_NUMA_NODES=0 to MPS wrapper script to avoid GPU NUMA nodes

[msimberg]

hwloc 2.11.0 started exposing the Grace-Hopper GPU NUMA nodes by default (https://github.com/open-mpi/hwloc/blob/41030697179b16f96f7e169f4530061c5fe6803f/NEWS#L164):

Version 2.11.0
--------------
...
* GPU support
  + Don't hide the GPU NUMA node on NVIDIA Grace Hopper.

This means that the MPS wrapper script would try to set GPU NUMA nodes to silly values (GPU NUMA nodes are indexed >= 4). For example:

$ hwloc-calc --physical --intersect NUMAnode $(hwloc-bind socket:0 -- hwloc-bind --get --taskset) # default
0,4
$ HWLOC_KEEP_NVIDIA_GPU_NUMA_NODES=0 hwloc-calc --physical --intersect NUMAnode $(hwloc-bind socket:0 -- hwloc-bind --get --taskset) # explicitly ignoring GPU NUMA nodes
0

This is not yet a problem with the system hwloc (on daint at least) which is at version 2.9.0, but if it's updated or if a newer hwloc is visible in an environment the MPS wrapper script would behave weirdly.

This PR explicitly sets HWLOC_KEEP_NVIDIA_GPU_NUMA_NODES=0 when getting the GPU index for a rank. Setting it doesn't hurt on older hwloc versions.

[github-actions]

preview available: https://docs.tds.cscs.ch/179

[RMeli]

LGTM, but maybe worth adding an annotation with a bit of explanation?

[msimberg]

LGTM, but maybe worth adding an annotation with a bit of explanation?

Yep, fair point. I added a comment in 2e406a2. What do you think?

[github-actions]

preview available: https://docs.tds.cscs.ch/179