Allow to modify procDriverNvidiaPath via ENV VAR or Flag

[ArangoGutierrez]

When working on NVIDIA/k8s-test-infra#182 , I can get the Mock to work with Device plugin as it only uses the NVML , but DRA uses NVML AND /proc/driver/nvidia/ and this path is hardcoded, not affected by the Helm nvidiaDriverRoot var .

k logs nvidia-dra-driver-gpu-kubelet-plugin-sbrdk -n nvidia-dra-driver-gpu -f
Defaulted container "compute-domains" out of: compute-domains, gpus, init-container (init)
I1016 09:23:29.655919       1 envvar.go:172] "Feature gate default state" feature="InOrderInformers" enabled=true
I1016 09:23:29.655977       1 envvar.go:172] "Feature gate default state" feature="InformerResourceVersion" enabled=false
I1016 09:23:29.655979       1 envvar.go:172] "Feature gate default state" feature="WatchListClient" enabled=false
I1016 09:23:29.655981       1 envvar.go:172] "Feature gate default state" feature="ClientsAllowCBOR" enabled=false
I1016 09:23:29.655983       1 envvar.go:172] "Feature gate default state" feature="ClientsPreferCBOR" enabled=false
I1016 09:23:29.667759       1 mount_linux.go:324] Detected umount with safe 'not mounted' behavior
Error: error creating driver: failed to create device library: error recursively unmounting /proc/driver/nvidia: failed to read directory `/proc/driver/nvidia:` open `/proc/driver/nvidia:` no such file or directory

If we could have

procDriverNvidiaPath = "/proc/driver/nvidia" 

Configurable via a flag / env var, then DRA runs on the MOCK System, and I even added IMEX support into the MOCK libnvidia-ml.so.1