Verify custom metrics fetch working and implement external metrics support

[Jeffwan]

🚀 Feature Description and Motivation

Our autoscaling framework already has multiple MetricFetcher implementations:

• RestMetricsFetcher → direct pod /metrics endpoint
• ResourceMetricsFetcher → Kubernetes resource metrics (cpu, memory)
• CustomMetricsFetcher → Kubernetes custom.metrics API
• ExternalMetricsFetcher → external sources (GPU-Optimizer, Kubernetes external.metrics)

Currently:

• Custom metrics fetcher is implemented, but needs end-to-end validation.
• External metrics fetcher has two paths:
• AIBrix GPU-Optimizer REST API (already wired through engineFetcher)
• Kubernetes external.metrics API (currently a TODO / not implemented).

We need to verify that custom metrics work in real environments and implement support for Kubernetes external.metrics API so operators can use external metrics (e.g., Prometheus Adapter) for autoscaling.

Use Case

Provide more metric source for users

Proposed Solution

No response