docs: add GPU Operator production troubleshooting runbook

The guide covers scenarios such as:
- GPUs not detected on nodes
- GPU workloads stuck in Pending
- driver daemonset failures
- device plugin initialization issues
- missing DCGM exporter metrics
- MIG configuration problems

It also includes example debugging commands and sample outputs to help
operators quickly recognize common failure patterns.

The content is based on past operational troubleshooting experiences running
GPU workloads on Kubernetes clusters using the GPU Operator.

Author: framsouza