Add local embedding option (Ollama) for Duplicate Issue Detector

[Skull-boy]

What

Add a local embedding option (e.g. Ollama's nomic-embed-text) as a free,
private alternative to the OpenAI embedding step in the
Duplicate Issue Detector workflow.

Why this matters

Right now the embedding step requires an OpenAI API key. A local option
would let anyone self-host this fully free and privately, matching the
pattern already used in the GitHub Debugger Agent workflow's LLM step.

Where

workflows/duplicate-issue-detector/workflow.json — the two
"Embed (OpenAI)" HTTP Request nodes.

Acceptance criteria

• Adds a documented, swappable local embedding node (Ollama)
• Updates the workflow's README with setup instructions for the local option
• Vector dimensions match whatever local model is used (note: local models
  often don't produce 1536-dim vectors like text-embedding-3-small —
  the Qdrant collection size may need to be reconfigured accordingly)

How to contribute

See CONTRIBUTING.md — branch off main, open a PR,
and I'll review before merging.

[mikemikimike]

Implemented in #28. Added opt-in Ollama nomic-embed-text nodes to both duplicate-detector flows, shared response extraction for OpenAI and Ollama vector formats, and documentation for endpoint configuration, 768-dimensional Qdrant collections, provider switching, and backfill. Local validation passed: 60 pytest tests, 5 contract validations, Ruff, compileall, JSON parsing, and wheel build. The repository CI run is action_required because GitHub requires maintainer approval for this fork workflow.