Add langchain-dkg integration

[haroldboom]

Integration submission: langchain-dkg

Bounty tag: cfi-dkgv10-r1
Tier target: Flagship

What this is

LangChain memory and retriever components backed by OriginTrail DKG v10. Gives LangChain agents durable, verifiable, queryable memory — conversation history is stored as Knowledge Assets in Working Memory and retrieved via tri-modal search (vector + SPARQL + text).

• PyPI: https://pypi.org/project/langchain-dkg/0.1.0/
• GitHub: https://github.com/haroldboom/dkg-langchain
• Design brief: https://github.com/haroldboom/dkg-langchain/blob/master/DESIGN_BRIEF.md
• Demo (Colab): https://colab.research.google.com/github/haroldboom/dkg-langchain/blob/master/demo.ipynb
• Demo (video): https://youtu.be/0XVYrikAZFQ

Components

Class	Purpose
DKGChatMessageHistory	LangChain BaseChatMessageHistory backed by DKG WM
DKGMemory	LCEL RunnableWithMessageHistory wrapper
DKGRetriever	BaseRetriever using SPARQL SELECT against WM/SWM

Submission checklist

• Entry file at integrations/langchain-dkg.json
• Slug follows normalization rules (langchain-dkg)
• install.kind is manual (Python library — pip install, not a CLI or service)
• memoryLayers declared: WM (default), SWM (via explicit promote)
• writeAuthority lists all mutating endpoints
• Curator operation (POST /api/assertion/{name}/promote) explicitly declared and explained in security.notes
• No external network egress (local DKG node only)
• No postinstall/install/preinstall scripts in the published package
• promotionPath filled in (WM → SWM → VM chain with UAL preservation)
• designBrief linked
• demo linked (interactive Colab notebook + video walkthrough)
• 25/25 unit tests passing
• MIT license consistent between registry entry and PyPI package
• commit pinned to exact SHA

[haroldboom]

Patch release 0.1.1 published.

While preparing for review I caught and fixed a few bugs in 0.1.0. The integration JSON in this PR is unaffected (no version pin), but anyone installing langchain-dkg via pip will now get the corrected build.

Fixes in v0.1.1:

• DKGRetriever — read SPARQL JSON results under the "results" key (per W3C spec) and extract .value from each binding cell. The previous code silently returned 0 documents against real DKG responses.
• DKGRetriever — escape SPARQL string literals to close a quote/backslash injection vector in the query argument.
• DKGRetriever — sync _get_relevant_documents now works inside Jupyter / async servers (was RuntimeError: This event loop is already running); added the run_manager kwarg required by BaseRetriever; migrated to Pydantic v2 ConfigDict.
• examples/research_agent.py — fixed an AttributeError (get_ual → get_turn_uri).
• DKGMemory — corrected docstring (.wrap → .wrap_chain, layer values wm/swm).

PyPI: https://pypi.org/project/langchain-dkg/0.1.1/
Diff: haroldboom/dkg-langchain@v0.1.0...v0.1.1

All 25 unit tests pass.