KNIX GPU monitoring/accounting capabilities

[ksatzke]

KNIX misses the capability to query the number of GPU devices and the GPU memory of devices in a particular deployment. However, this functionality is required when configuring a KNIX microfunctions workflow using a GPU to the platform, because in contrast to CPU or memory resources, GPU resources cannot be oversubscribed.

For this purpose, the total available GPU memory (quantity * memory) of each cluster node, in addition to the number of GPU devices on the node, needs to be reported.