For pm-gpu: add gpu affinity flag to srun and add pelayouts for coupled ne256 cases

[ndkeen]

For pm-gpu, add option to srun that sets GPU affinity with a new shell script, but only if MPI's per node is 64 or larger.
Which might only happen with certain pelayouts designed to use new xstrid option, such as these in this PR which
add S/M/L pelayouts for ne256-wcyclxx.
Remove special GPU bind case for MMF compsets which are no longer being used.

These are changes from #7818

BFB

[ndkeen]

I verified we still get same performance using the GPU affinity setting even when it is always used (ie, not just when using 64 or more MPI's per node). However, with MPS, I had a case that was a solid 2% slower with using GPU affinity. We can use less than 64 MPI's per node with MPS, or find another way to turn it off if this was ever needed. Currently, MPS is only used for testing.

I also tested the branch with all GPU suites I could think of:

e3sm_eamxx_v1, e3sm_eamxx_large, e3sm_eamxx_extra_large, e3sm_gpuacc, e3sm_gpucxx

[rljacob]

These look like the same changes as in 7818 so redo this be either cherry picking commits form there or use --author option in the git commit to add Az's authorship.

[ndkeen]

Yes these are the same changes in the other PR, but I was running into issues testing in the older branch. It may only need a rebase, but this was easiest/quickest path for me.

[ndkeen]

how best to proceed?

[rljacob]

Close this and redo the branch using --author on the commit so Az is the author. Or Az can do it. Or just rebase Az's original branch and merge that. What exactly was the problem you had working with it?

[ndkeen]

I dont recall the issues now, but it was messy dealing with branch.
Can Az rebase?