Kyma Telemetry auto-wiring

When Kyma Telemetry module is present on the cluster, automatically create a MetricPipeline CR that scrapes DCGM metrics from the gpu-operator namespace and forwards them to the user's existing metrics backend.

NVIDIA DCGM Exporter already ships GPU metrics (utilization, memory, temperature) inside the gpu-operator namespace. Today they go nowhere. With this feature, enabling the GPU module is enough - metrics appear in the user's dashboards automatically with zero additional configuration.

How

1. On every reconcile, check if MetricPipeline CRD exists on the cluster
2. If Telemetry is present, create a MetricPipeline pointing at gpu-operator namespace, forwarding to Kyma's internal OTel Collector
3. Delete the pipeline when GPU module is deleted

Open question

Confirm the internal Telemetry OTel Collector service endpoint - this is what we point the pipeline output at so metrics flow to whatever backend the user already configured in Telemetry.

Acceptance criteria

• Telemetry not installed -> no error, no pipeline, silent skip
• Telemetry installed -> MetricPipeline created automatically on GPU module install
• GPU module deleted -> MetricPipeline cleaned up
• Telemetry installed after GPU module -> next reconcile picks it up
• GPU metrics visible in user's backend with no manual steps

Start by tracing the GPU module's reconcile flow and deletion handling, then inspect how the MetricPipeline CRD and Kyma Telemetry's internal OTel Collector endpoint are discovered. Confirm the pipeline targets the gpu-operator namespace and is created, skipped, or removed across the installation and deletion scenarios; done means all listed acceptance criteria pass.