doc: add doc best-practice/configuring-rolling-updates

[diw-zw]

No description provided.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[cheyang]

/copilot review

[cheyang]

@gemini-code-assist /review

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[cheyang]

Solid documentation PR adding rolling update best-practice guides in both English and Chinese. Each language gets a concept doc (explaining parameters, rationale, and architecture) and a hands-on operations guide (step-by-step kubectl workflows with expected outputs). Content is technically accurate and well-structured.

Two minor nits:

1. Typo in doc/best-practice/zh/03-configuring-rolling-updates-guide.md: In the CoordinatedPolicy verification output, 50mm should be 50m.
2. Dangling links: The "Related Documents" sections in both concept docs reference 01-deploy-inference-service.md, 02-using-role-templates.md, and 08-configuring-autoscaling.md which do not exist yet.

Neither issue is blocking. LGTM overall.

[cheyang]

The documentation examples all use lmsysorg/sglang:v0.5.9 (10GB+) but the actual command is just sleep 3600 for control-plane validation. This makes it impractical for users who want to quickly follow along — pulling a multi-GB inference image just to run sleep is heavy.

Could you swap the demo image to something lightweight like busybox:1.36 or nginx:1.25-alpine? The rolling update behavior being demonstrated doesn't depend on the inference runtime, so a small image would make the guide much more accessible for testing.

[cheyang]

One more user-experience note (separate from the image-size one above):

CoordinatedPolicy provenance is unclear. Operation 4 introduces a CoordinatedPolicy CRD but doesn't say whether it comes with the RBG Controller install or requires a separate step. First-time users who follow the earlier docs and then hit this one may get "no matches for kind CoordinatedPolicy" and not know how to recover. A one-line "This CRD is installed together with the RBG Controller via the same Helm chart / kustomize" (or the opposite) would fully unblock this.

Also: the kubectl get pods -w output in Operation 4 shows some Pods entering Error state briefly. That's actually normal — sleep receives SIGTERM and exits non-zero — but the doc doesn't say so, and a user watching this stream will assume something is broken. A short annotation like # Error is normal: sleep exits non-zero on SIGTERM would prevent panic.