Moving from `pynvml` to `nvidia-ml-py`

[jakirkham]

Currently this is making use of pynvml in a few places:

benchmark/rapids_pytest_benchmark/setup.py

Line 22 in f126064

install_requires=["pytest-benchmark", "asvdb", "pynvml", "rmm"],

benchmark/parser/GBenchToASV.py

Line 11 in f126064

from pynvml import smi

However we would like to move to nvidia-ml-py in the future. Raising this issue to track this work