Spike: verify Foundry Local embedding-model support for RAG

[jongalloway]

Summary

The workshop is a RAG app needing both a chat model and an embedding model (semantic search). Foundry Local is focused on generative/chat SLMs; embedding-model coverage needs verification before we document it as a complete fallback.

Tasks

• Determine whether Foundry Local can serve an embedding model usable by IEmbeddingGenerator
• If yes: document the model + wiring, and Foundry Local becomes a full-RAG offline path
• If no: document Foundry Local as chat-only, and keep Ollama (e.g., all-minilm) as the full-RAG offline path
• Record the finding in the alternative-providers issue/section

Acceptance criteria

• A definitive answer on Foundry Local embeddings, with the offline-provider docs updated accordingly.

[jongalloway]

Blocker for the manual-RAG part (#500). The hand-written RAG exercise needs an embedding generator, so this spike decides the offline story: if Foundry Local can't serve embeddings, Ollama all-minilm becomes the offline full-RAG path. Resolve early.