Part 5: providers & fallbacks walkthrough (#502)

[jongalloway]

Summary

Adds the new Part 5: Providers & fallbacks — the dedicated home for the "swap the provider, same IChatClient" payoff and the GitHub Models retirement story. Resolves #502 (absorbs #495).

What's included

• Part 5 - Providers and Fallbacks/README.md — a walkthrough covering:
  • Azure AI Foundry as the primary provider (recap of Parts 2–4)
  • The universal OpenAI-compatible pattern (OpenAIClient + endpoint + key) that reaches every other provider with the same three lines
  • Foundry Local — offline chat (SLMs), with an embeddings-support note pointing at the Spike: verify Foundry Local embedding-model support for RAG #496 spike
  • Ollama — fully-offline RAG (chat + all-minilm embeddings)
  • GitHub Models (legacy) with a prominent retirement warning (retires 2026-07-30, brownouts July 16 & 23)
  • A provider comparison table and a "what changed = config, not code" summary

Accuracy

• Foundry Local (Microsoft.AI.Foundry.Local + OpenAI, FoundryLocalManager → local OpenAI-compatible web service) and Ollama (http://localhost:11434/v1) patterns were verified against current Microsoft Learn docs.
• Reinforces the core theme: all non-Azure providers use the same OpenAIClient → .AsIChatClient(), so switching is a registration change.

Notes

• Docs-only — no new buildable project; the snippets slot into the Part 2/3 projects.
• Foundry Local is documented as the local chat path; fully-offline RAG uses Ollama (pending the Spike: verify Foundry Local embedding-model support for RAG #496 embeddings decision).