Add gpu count metrics including limit, request and total counts

[movence]

** Revision **

• Remove the logic to copy resource metric attributes down to data points

Description:
Add NVDIA GPU count metrics including _limit, _request and _total at pod, node and cluster levels. This change will have the leader agent pod to collect GPU count metrics rather than individual agent pods to collect them. This is to collect limit and request metrics for pods that are still in pending status due to lack of available GPU devices. The leader agent will still gather information of pending pods for limit and request, but *_total metric will only include gpu counts data from running pods.

Testing:
Tested with a cluster which has 2 g4dn.12xlarge instances with 4 gpu devices each. There are 4 RUNNING pods (total 8 allocated gpu devices) and 2 PENDING pods requesting 2 gpu devices each.