Add new constant called SYSTEM_PROCESS_NAME_BUFFER_LENGTH

[klueska]

Previously we were using the value of C.PR_SET_NAME to limit the size of
the buffer used to write the process name into when calling
'nvmlSystemGetProcessName()'. While this technically wouldn't cause a
buffer overrun, it limited the length of the process name by a
(somewhat) arbitrary amount since C.PR_SET_NAME is not meant to indicate
a maximum length for a process name, but rather an arbitrary code for
telling 'prctl()' to set the name for a process (which has nothing to do
with its length).

It was a clear oversight to use this random value as the buffer length.
In the new implementation we simply hard code a new variable called
SYSTEM_PROCESS_NAME_BUFFER_LENGTH to 256 (which is well more than the
actual maximum process name length on linux). Having a larger value
won't hurt anything, and will make it more portable if / when we decide
to start supporting windows.

[elezar]

Thanks @klueska . One minor nit on the naming. Other nvml functions have constants such as NVML_SYSTEM_NVML_VERSION_BUFFER_SIZE, so we could consider something like:

NVML_SYSTEM_NVML_PROCESS_NAME_BUFFER_SIZE

instead of the quite generic PROC_NAME_MAX. Not a blocker though.

[klueska]

I was actually trying to make it reminiscent of PATH_MAX on linux. However, now that you point it out, I actually think it should just be an internal variable and not exported at all. It’s definitely not something that should be namespaced with NVML since it’s not part of the original NVML namespace, and it’s just an internal implementation detail at this point.