Fix duplicate exported_job labels with transform processor

[samuelarogbonlo]

Description

This PR fixes an issue where Prometheus metrics (e.g., target_info) contain exported_job labels with namespace prefixes (e.g., production/my-app). This label change began after upgrading to OpenTelemetry Operator v0.125.0 and Collector v0.126.0, causing duplicate time series in Prometheus.

The solution introduces a transform processor that strips namespace prefixes from exported_job labels, ensuring consistent metrics and compatibility with existing dashboards and alerts.

It provides:

1. A Dockerfile for building a custom OpenTelemetry Collector image
2. A collector-config.yaml with the transform processor included
3. A kubernetes-deployment.yaml example manifest for maintainers

Link to tracking issue

Fixes #13079

Testing

1. Local testing with custom-built collector:

• Confirmed that metrics previously labeled as production/my-app are now my-app
• Verified Prometheus target and label consistency:
  curl http://localhost:8889/metrics
  Confirmed no duplicate exported_job label values.

2. Health check verified:
   curl http://localhost:13133
{"status":"Server available","upSince":"...","uptime":"..."}
   Image builds and runs cleanly:

• Custom collector compiled and runs with no config errors
• No unintended metrics dropped

Documentation

This PR includes:

• Dockerfile: Custom OpenTelemetry Collector image with the transform processor
• collector-config.yaml: Configuration demonstrating the namespace-stripping fix
• kubernetes-deployment.yaml: Example deployment manifest for maintainers

Detailed deployment and testing instructions are in this PR description for maintainers to implement or adapt as needed.

Notes

This solution provides maximum flexibility:

• Choose the deployment option that best fits infrastructure
• No changes required to existing Prometheus or dashboard configurations
• Can be deployed immediately or integrated into existing collector configs
• Comprehensive testing instructions included

The transform processor approach is the least disruptive solution that maintains full backward compatibility while solving the duplicate time series issue.