feat(ray-on-eks): batch LLM inference blueprint with Iceberg + tokenomics

[vara-bonthu]

What does this PR do?

🛑 Please open an issue first to discuss any significant work and flesh out details/direction. When we triage the issues, we will add labels to the issue like "Enhancement", "Bug" which should indicate to you that this issue can be worked on and we are looking forward to your PR. We would hate for your time to be wasted.
Consult the CONTRIBUTING guide for submitting pull-requests.

A complete, runnable batch LLM inference blueprint on the ray-on-eks stack. DeepSeek-R1-8B on vLLM 0.22 / Ray 2.56 (KubeRay v1.6), triaging mock support tickets into structured output (sentiment, urgency, category, summary).

• Ships an end-to-end example (data-stacks/ray-on-eks/examples/ray-batch-inference/) with a one-command deploy.sh: stage weights to S3, mirror the image to ECR, generate data, and run inference.
• Implements both batch inference patterns against the same dataset so the tradeoff is measurable. a shared RayService endpoint (pattern 1) and the native ray.data.llm engine-in-job (pattern 2).
• Lands tickets in an Apache Iceberg table (Glue catalog) and reads them incrementally via a per-pattern watermark, so re-runs only process new rows (at-least-once, idempotent output).
• Adds a Grafana dashboard + PodMonitor (DCGM GPU cost/utilization + Ray Data pipeline throughput) wired into the stack's kube-prometheus, all open-source.
• Enables enable_nvidia_gpu_operator, SOCI FastImagePull + RAID0 on the GPU node class, and weight streaming from S3 via the Run:ai Model Streamer.
• Publishes a full website guide with measured cold-start anatomy, a tokenomics cost comparison, and architecture diagrams.

Motivation

• Teams that need both real-time and batch inference usually build two separate stacks; this shows Ray collapsing them onto one platform on EKS.
• Batch on ephemeral GPUs is the cost-efficient pattern (rent the GPU only while the job runs) but only works if cold start is fast. the blueprint solves that with same-region ECR + S3 weight streaming.

More

• Yes, I have tested the PR using my local account setup (Provide any test evidence report under Additional Notes)
• Mandatory for new blueprints. Yes, I have added a example to support my blueprint PR
• Mandatory for new blueprints. Yes, I have updated the website/docs or website/blog section for this feature
• Yes, I ran pre-commit run -a with this PR. Link for installing pre-commit locally

For Moderators

• E2E Test successfully complete before merge?

Additional Notes