docs: add runner in-cluster vcluster access pattern guide

[weicao]

Summary

New methodology doc addon-runner-incluster-vcluster-access-pattern-guide.md covering when addon tests on a remote IDC vcluster, the runner pod should sit inside the host k8s cluster and reach the vcluster API via in-cluster service DNS, taking the workstation network out of the test critical path.

Body (generic methodology, version-agnostic, no engine binding):

• §1: 3 vcluster API access topologies — workstation+port-forward / workstation+NodePort / runner-pod+in-cluster-DNS — with their respective failure classes
• §2: 5 hard rules — kubeconfig server field = in-cluster service DNS; long-running pod (sleep infinity); nohup ... & so kubectl exec returns immediately; evidence retrieval via kubectl exec -- tar c piped to tar x (avoids kubectl cp etcd timeout); closeout treats host-API impact as independent from runner result
• §3: 5-point PR / design review checklist
• §4: 2 anti-pattern vs correct-pattern pairs
• §5: relation to addon-host-runner-job-pattern-guide.md, addon-test-runner-cadence-discipline-guide.md, addon-test-acceptance-and-first-blocker-guide.md, addon-evidence-discipline-guide.md

Appendix A is OceanBase enterprise addon migration from workstation + kubectl port-forward to idc4 host-runner + in-cluster service DNS, plus reverse example showing the prior path's pf-staleness T1_vcluster_api_unreachable_via_macD_portforward_mid_run blocker. Explicit boundary: 3-sample observation supports only "no workstation-network impact on in-pod test execution in those samples"; NOT extrapolated to permanent immunity.

SKILL-INDEX.md updated: added entry under ### 5. 改造 runner / 工具链.

Test plan

• Manual: methodology body has zero OB-specific commands
• Manual: appendix is the only OB-specific section and has explicit boundary
• Manual: cross-doc references resolve

🤖 Generated with Claude Code

[weicao]

Blocking for merge:

1. PR body still contains 🤖 Generated with [Claude Code].... Public PR body must not include AI/tool attribution.
2. The new guide intro has only 4 standard fields. Please add > **Affected by version skew**: ... after Applies to KB version.
3. The appendix references addon-runner-portforward-staleness-keepalive-guide.md as a sibling that is only in PR docs: add runner port-forward staleness keepalive (stop-gap) guide #126, not on main. Either remove/defer that reference, or merge/rebase in dependency order after the target exists.

The structural recommendation itself looks good: host-runner + in-cluster DNS removes workstation network from the test critical path.

[weicao]

Closing per Allen's docs PR queue triage (DM 2026-05-18). Superseded by main-branch standard guide(s).