feat: update charts for enabling monitoring

[Fei-Guo]

Reason for Change:

Adds GPU and inference observability wiring for Azure Monitor managed Prometheus across the two chart layers:

• charts/modelharness (per-namespace): new templates/azure/podmonitor.yaml renders an azmonitoring.coreos.com/v1 PodMonitor that scrapes every pod in the workload namespace across the EPP, Istio gateway, and vLLM metric surfaces, lifting the inferencepool / gateway / shadow-pod / workspace pod labels onto the emitted series. Gated on cloudprovider=azure + a new azure.podMonitor toggle, and additionally on .Capabilities.APIVersions.Has "azmonitoring.coreos.com/v1/PodMonitor" so an install against a cluster without the Azure Managed Prometheus add-on silently skips the CR instead of failing on an unknown kind. Values and values.schema.json updated (enabled / name / interval).
• charts/productionstack (cluster-wide): new templates/azure/ama-metrics-settings-configmap.yaml renders the add-on's well-known ama-metrics-settings-configmap in kube-system enabling the dcgmexporter default scrape target, so the add-on collects raw NVIDIA DCGM GPU metrics cluster-wide. Gated on cloudprovider=azure via a new azure.amaMetricsSettings toggle (default true; renders nothing on non-Azure installs).

This gives per-namespace inference metrics plus cluster-wide raw GPU metrics in Azure Monitor managed Prometheus, with both templates safe to apply on clusters where the add-on / CRD is absent.

Requirements

• added unit tests and e2e tests (if applicable).

Notes for Reviewers:

• PodMonitor label caps (labelLimit: 63, labelNameLengthLimit: 511, labelValueLengthLimit: 1023) are hand-picked guardrails, not Azure/Prometheus defaults (Prometheus default is 0 = unlimited). Flag if you'd prefer them tunable or dropped.
• ConfigMap ownership: productionstack now owns ama-metrics-settings-configmap. If a cluster already has one (platform- or operator-managed), helm install will conflict — set azure.amaMetricsSettings.enabled=false there. Default is true.
• .Capabilities caveat: capability discovery needs a live API server, so a bare helm template (no --validate/--api-versions) will not render the PodMonitor even on a cluster that has the CRD. Relevant for CI snapshot tests.
• No values.schema.json exists for charts/productionstack, so only charts/modelharness's schema was updated.
• The modelharness podMonitor.name default is all-model-pods; the ama settings ConfigMap name/namespace are pinned (the add-on only reads that exact name in kube-system).