What type of PR is this?
/kind feature
/kind test

What this PR does / why we need it:

Extracts the vllm-render container from the EPP Deployment sidecar into its own standalone Deployment + Service. This makes the render service independently scalable — operators can tune replica count without touching the EPP deployment. It also removes the tight coupling between EPP and render lifecycle.

For e2e tests, replaces the real vllm-render instance with the inference simulator image running with --force-dummy-tokenizer.

This eliminates:

• the HuggingFace API call on every test-case startup (which required network access and hit an external dependency unrelated to what routing tests validate)
• the per-test render container startup wait
• the runtime YAML mutation logic (removeRenderSidecar, isModelReal, removePodSpecListItem) that patched out the sidecar for simulated-model tests

The EPP config is updated to point at http://vllm-render:8082 (the new Service) instead of http://localhost:8000, and simulators receive --force-dummy-tokenizer to match.

Which issue(s) this PR fixes:

Ref #1595

Release note (write NONE if no user-facing change):

vllm-render is now deployed as an independent Deployment and Service rather than a sidecar in the EPP pod, allowing it to be scaled independently.