AKS E2E Clouds placement audits fail on nvsentinel's NIC health monitor #567

What happened?

The AKS leg of E2E Clouds fails on every PR that touches the serving stack. The placement audits flag nvsentinel's NIC health monitor:

FAILED e2e/clouds/test_placement.py::test_tolerations_are_scoped
  'nvsentinel/nic-health-monitor-bhtnq tolerates the GPU taint'
  'nvsentinel/nic-health-monitor-jm487 tolerates the GPU taint'
FAILED e2e/clouds/test_placement.py::test_only_gpu_workloads_on_the_gpu_pool
  "nvsentinel/nic-health-monitor-bhtnq doesn't belong on fake GPU node kwok-gpu-1"
  "nvsentinel/nic-health-monitor-jm487 doesn't belong on fake GPU node kwok-gpu-0"

#539 enabled the NIC health monitor in the AKS stack, and only there:

Like nvsentinel's other health monitors, it belongs on GPU nodes and tolerates the GPU taint. But it isn't in GPU_WORKLOADS or in AKS's gpu_workloads. The audits treat any pod missing from both as misplaced:

E2E Clouds only runs on PRs that touch functions/compose-serving-stack/** or e2e/**, so main never ran it after #539 merged. #542 is the first PR to hit it (run).

#539's own AKS run passed, probably because the audits race nvsentinel's labeler. #569 tracks that.

How can we reproduce it?

1. On main, run nix run .#e2e -- --cloud aks.

What environment did it happen in?

Modelplane version: main (69ae1d7)

nvsentinel v1.25.0, on the KWOK-faked AKS leg of E2E Clouds.