Backport of CNV-74367 fix to release-4.18.

The original fix was merged to release-4.19 via PR #3256, but automatic cherry-pick to release-4.18 failed due to significant codebase differences.

Root Cause:
For CPU requested metric, we were fetching from Prometheus using kube_pod_resource_request which is not directly related to the VMI CPU requested - it's scaled using HCO configuration.

Fix:
Get CPU requested from the VMI spec (sockets × cores × threads) instead of Prometheus query.

Original ticket: CNV-74367
Original PR: https://github.com/kubevirt-ui/kubevirt-plugin/pull/3256