[cfi-dkgv10-r1] Add github-dkg: GitHub knowledge → DKG v10 Working Memory

[haroldboom]

What this PR does

Adds github-dkg as a community integration. Ingests GitHub issues, pull requests, and review comments into DKG v10 Working Memory as Knowledge Assets, with full provenance and selectable promotion to Shared Working Memory.

Integration links

• Repo: https://github.com/haroldboom/github-dkg
• Commit pinned: f2b9f67c9518137e5fc197507205520f3e402c71
• Published package: github-dkg@0.1.0
• Design brief: https://github.com/haroldboom/github-dkg/blob/master/DESIGN_BRIEF.md
• Demo (3-min walkthrough): https://youtu.be/9Tgb4-OXGuk
• Colab notebook (offline mocks): https://colab.research.google.com/github/haroldboom/github-dkg/blob/master/demo.ipynb

What it does

• CLI (github-dkg ingest|ingest-one|search|promote) for one-shot or bulk ingestion of any repo's issues / PRs / reviews. --since supported on both endpoints (server-side for issues; sort=updated&direction=desc + early-stop for PRs since /pulls has no native since).
• GitHub Action (Docker, fires on issues / pull_request / pull_request_review) for automatic per-event ingestion. The bundled examples/workflow.yml shows the canonical SWM-promotion gate: only PRs labelled architecture-decision and merged to main.
• Each item becomes one tri-modal Knowledge Asset (markdown → structural triples + semantic embedding) via POST /api/memory/turn.
• Promotion to SWM via POST /api/assertion/{name}/promote is always explicit, never automatic.

Security highlights

• Network egress: only api.github.com and the local DKG node.
• No PUBLISH (Verified Memory) calls.
• GitHub rate-limit responses (403/429) surface as a typed GitHubRateLimitError rather than crash mid-traversal.
• No postinstall/preinstall scripts; hatchling-built wheel uploaded to PyPI via twine.
• 23/23 unit tests pass against the published wheel (verified in a fresh venv outside the source tree).

Sister submission

langchain-dkg — PR #4 is the same author's other Round 1 entry — a LangChain memory adapter backed by DKG v10. Together they form a write-side / read-side pair: github-dkg writes engineering tacit knowledge into the substrate, langchain-dkg gives any LangChain agent persistent memory backed by it.

Submission tag: cfi-dkgv10-r1