policy setup latency duration should be per pod

[yash97]

Issue #, if available:

Description of changes:

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[viveksb007]

why it should be per pod?

having per pod observability is great but currently its telling how much time it took whenever a reconciliation request came for it. whenever a PE reconcile comes, it fetches all sibling PE(s) and then processes as a complete Network Policy.

[yash97]

Current metric is ambiguous: we Observe() inside the loop with a single start, so each iteration includes all prior work. The 2nd pod’s “latency” = derive targets + pod1 setup + pod2 setup, etc. That’s neither true “per-reconcile” nor true “per-pod”.

We have two options:

Option A (preferred): move Observe() outside the loop → one value = total policy setup time for all affected pods.

Option B: time each pod with its own start and observe per-pod (and optionally add a separate counter for pods processed). (What this PR does, but now I think Option A makes more sense.)

[viveksb007]

agreed current metric is ambiguous and doesn't provide good signal.

Option A makes sense to me.

today, this latency can be impacted by other NP(s) which target the pods which this current NP is targetting.