OpenShift: improve deployment reliability and egress access

[TomasTomecek]

Summary

Improve OpenShift deployment reliability and enable LLM access via Vertex AI.

Changes

Egress & Networking:

• Allow egress to oauth2.googleapis.com, vertexai.googleapis.com, and googleapis.com for Vertex AI/Claude authentication

Deployment Script:

• Add retry logic to oc import-image with exponential backoff (2s, 4s, 8s, 16s delays, max 5 attempts) to handle transient OpenShift conflicts

Documentation:

• Update OpenShift README with deployment location (jotnar-ymir--jotnar-ymir on GPC cluster), list of all deployed components, and current deployment procedure

[gemini-code-assist]

Code Review

This pull request updates the OpenShift deployment configuration by introducing a new deployment script with retry logic, enabling the Phoenix observability stack, and renaming the DRY_RUN environment variable to SILENT_RUN. It also updates the README with deployment locations and adds egress rules for Google OAuth and Vertex AI. Feedback includes addressing shell portability issues in the deployment script, removing redundant namespace flags, and ensuring the retry logic matches the intended exponential backoff. Additionally, it is recommended to reference environment variables from ConfigMaps for better maintainability and to refine broad egress rules for Google APIs.

[gemini-code-assist]

The implementation uses a fixed 2-second delay, which contradicts the 'exponential backoff' mentioned in the pull request description. If exponential backoff is intended, the sleep duration should increase with each retry attempt.