docs: small-batch reconciliation of blog draft content into repo docs

[resker]

Summary

Folds high-value content from an unpublished blog draft into the repository docs so the blog can reference canonical sources rather than being the sole host of this material. All four changes are additive, independent, and low-risk.

Changes

• README.md — expand the "Motivation and Problem Statement" section (merged via docs(readme): add 'Motivation and Problem Statement' section #247) with the disaggregated inference scenario: prefill/decode separation, KV cache transfer across the worker boundary, and the NVLink vs. Ethernet performance cliff. Complements the existing distributed-training framing.

• docs/providers/infiniband.md — add a "Why automate IB discovery?" paragraph with the scale argument: hand-maintained labels work at ~32 nodes, break at 1,000 with fabric churn.

• docs/providers/netq.md — add an "Observed vs. Intended Topology" section describing NetQ's distinctive ability to observe degraded-but-up links via live telemetry. This signal is invisible to ibnetdiscover-based discovery and unreported by any cloud placement API.

• CONTRIBUTING.md — expand beyond the current 62 lines of DCO boilerplate with:

  • Community section — Kubernetes Slack channels (#topology-aware-scheduling, #gpu-nvidia)
  • Areas where contributions are especially welcome — new providers, on-premises network fabrics, Kubernetes Workload API / KEP-5732, KEP-4962 upstream label schema, Grove ClusterTopology integration

Deferred to a follow-up

One additional small-batch item from the same blog-gap analysis — a KEP-4962 "Upstream alignment" paragraph in docs/reference/node-labels.md — is intentionally deferred to a follow-up PR, since that file does not yet exist on main (it's introduced by #254).

Test plan

• Markdown renders cleanly in GitHub preview for all four files
• All external links verified (KEP-5732, KEP-4962, Grove, Kubernetes Workload API, Kubernetes Slack channel archives)
• Provider list in CONTRIBUTING.md (AWS, GCP, OCI, Nebius, Lambda AI, CoreWeave) verified against pkg/registry/registry.go
• No conflicts with other open PRs — this PR touches README, infiniband.md, and netq.md in sections that do not overlap with the edits in open PR docs: document gpu.clique relationship and non-MNNVL topology source #255

[greptile-apps]

Greptile Summary

Four additive, independent documentation changes that fold content from an unpublished blog draft into the repo: a disaggregated-inference paragraph in README.md, a scale-motivation paragraph in docs/providers/infiniband.md, an "Observed vs. Intended Topology" section in docs/providers/netq.md, and a Community section with Kubernetes Slack links in CONTRIBUTING.md. No code is changed.

Confidence Score: 5/5

Safe to merge — docs-only, no code changes, all four additions are technically accurate and well-placed.

All changes are additive documentation with no code modifications. Content is technically sound and consistent with the existing provider docs. No broken links introduced in the changed content, and no P0/P1 findings.

No files require special attention.

Important Files Changed

Filename	Overview
README.md	Adds a disaggregated inference paragraph to the Motivation section, illustrating KV-cache transfer cost and NVLink vs. Ethernet performance cliff; technically accurate and well-integrated into the existing narrative.
CONTRIBUTING.md	Adds a Community section with Kubernetes Slack links and a reference to the pinned Roadmap issue; expansion is clean and doesn't introduce any broken references in the current content.
docs/providers/infiniband.md	Prepends a "Why automate IB discovery?" paragraph giving a concrete scale argument (~32 vs 1,000 nodes); fits naturally before the existing provider choice guidance.
docs/providers/netq.md	Inserts an "Observed vs. Intended Topology" section highlighting NetQ's live telemetry advantage over ibnetdiscover or static labels; placed correctly before the Output section.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Cluster topology needed] --> B{Fabric type?}
    B -->|Cloud: AWS / GCP / OCI / Nebius| C[CSP provider]
    B -->|Spectrum-X or MNNVL infra| D{NetQ deployed?}
    B -->|InfiniBand, no NetQ| E{Deployment?}

    D -->|Yes| F[NetQ provider\nobserved live topology]
    D -->|No| E

    E -->|Bare-metal / Slurm| G[infiniband-bm\nibnetdiscover via pdsh]
    E -->|Kubernetes| H[infiniband-k8s\nibnetdiscover via pod exec]

    F --> I{Also need K8s MNNVL scheduling?}
    I -->|Yes| J[DRA provider\ncoexists with NetQ]
    I -->|No| K[Engine: slurm / k8s / slinky]
    G --> K
    H --> K
    C --> K

Loading

_{Reviews (2): Last reviewed commit: "docs: reconcile small-batch gaps from bl..." | Re-trigger Greptile}

[greptile-apps]

Broken internal link to non-existent AGENTS.md

./AGENTS.md is referenced as the canonical guide for adding a new provider, but the file does not exist anywhere in the repository (confirmed via full tree search). Contributors following this link will hit a 404. Either this file needs to be created before (or in) this PR, or the reference should point to an existing doc that describes the provider interface.

Suggested change

	- **Additional cloud and colocation providers.** If your infrastructure isn't covered by the existing AWS, GCP, OCI, Nebius, Lambda AI, and CoreWeave providers, adding support means integrating with your provider's topology API and implementing a new Topograph provider. The provider interface is documented in the repository's [`AGENTS.md`](./AGENTS.md) "Adding a New Provider" section.
	- **Additional cloud and colocation providers.** If your infrastructure isn't covered by the existing AWS, GCP, OCI, Nebius, Lambda AI, and CoreWeave providers, adding support means integrating with your provider's topology API and implementing a new Topograph provider. The provider interface is documented in the repository's provider source under `pkg/providers/`.