* Add GPU cost overview dashboards

Adds 5 dashboards to the GPU integration for monitoring GPU compute
spend and utilization across cloud providers and Kubernetes.

Dashboards:
- gpu_cost_overview: cross-cloud totals, spend by team/env, fleet utilization
- aws_gpu_cost_overview: AWS-specific with Capacity Block tracking
- azure_gpu_cost_overview: Azure GPU VM families (NC, NCv3, ND series)
- gcp_gpu_cost_overview: GCP GPU SKUs with On-Demand vs Committed coverage
- k8s_gpu_cost_overview: cluster/namespace allocation, idle cost attribution

Cost queries span cloud_cost amortized metrics plus unblended for
AWS Capacity Blocks (which are not captured in amortized). Utilization
widgets join GPU telemetry (gpu.sm_active, gpu.device.total) with
cost data for unit-economics views.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Refine GPU utilization metrics and dashboard polish

- Switch unhealthy KPIs to GPU Idle % (gr_engine_active) since
  gpu.device.unhealthy is non-functional
- Add Healthy GPU Rate KPI on k8s using kubernetes_state.node.gpu_allocatable
  / gpu_capacity
- Add Spend on Idle GPUs KPI to per-cloud and overview dashboards
  (excludes AWS Capacity Blocks since their cost is upfront and unrelated
  to engine activity)
- Standardize utilization terminology: "Average GPU Utilization %" and
  "GPU Idle %" across dashboards
- Drop redundant cloud provider prefix from per-cloud widget titles and
  remove the "GPU Spend" group wrapper to match k8s dashboard layout

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* Add GPU Monitoring setup CTA banner to per-cloud and k8s dashboards

Mirrors the existing banner from the overview dashboard so users on any
cloud-specific or Kubernetes view see the link to enable GPU Monitoring,
which populates utilization-driven widgets.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…23609)

* fix(nutanix): normalize all ntnx_* tags to lowercase with $unknown fallback

Route every enum-backed ntnx_* tag (host_type, hypervisor_type,
node_status, plus the previously-handled state tags) through a single
_norm_state helper so they all follow one rule: lowercase the API
value, fall back to "\$unknown" when the source is missing.

Picks "\$unknown" (the API spec's own sentinel) as the fallback so
there's no mismatch between "value present but says \$UNKNOWN" and
"value missing" — both surface as ntnx_X:\$unknown.

ntnx_disk_status's "unknown" fallback is updated to "\$unknown" for
the same reason.

* docs(nutanix): add changelog for tag normalization

* docs(nutanix): shorten changelog to one customer-facing line

* refactor(nutanix): extract tag values into named variables

Hoist _norm_state and get_nested calls out of f-strings in the
tag-extraction helpers. Each tag computation now binds to a named
local first, making the read top-down and easier to step through.

* refactor(nutanix): rename _norm_state to _normalize_tag_value

The helper is used for type, state, mode, and status tags — not just
state — so the broader name better describes what it does.

* refactor(nutanix): collapse node_status to one variable, restore tags = []

Lowercase the node-status comparison sets so the normalized tag value
serves both the status_value lookup and the tag emission, removing the
need for a separate node_status_tag local.

Restore the tags = [] preamble in the tag-extraction helpers since it
makes the building intent obvious.

* fix(nutanix): normalize powerState in vm.status gauge

Match what _report_host_status_metrics does: route the powerState
lookup through _normalize_tag_value and lowercase the comparison
literals. Removes the asymmetry where vm.status was the only metric
still relying on raw uppercase API values.

Addresses review feedback on PR #23609.

* refactor(nutanix): normalize disk statuses in _aggregate_disk_status

Lowercase DEGRADED_DISK_STATUSES at the constant and route disk
``status`` values through ``_normalize_tag_value`` so the comparison
surface matches the rest of the module. Aligns with the convention
introduced in ``_report_host_status_metrics``.

* test(nutanix): cover \$unknown fallback for host enum tags

Verify ``ntnx_host_type``, ``ntnx_hypervisor_type``, and
``ntnx_node_status`` emit ``\$unknown`` when ``hostType``,
``hypervisor.type``, and ``nodeStatus`` are absent from the host
payload, the always-emit behavior introduced in this PR.

* test(nutanix): align unknown-fallback test with repo conventions

delete 'pause_auto_refresh'