Add Ollama support

[patmakesapps]

Summary

• add wrapOllama support for chat, generate, and embed-style usage tracking
• extend the fetch proxy to recognize Ollama local/cloud API endpoints and parse NDJSON streaming responses
• add README docs, focused Ollama tests, and a one-command local smoke test via npm run ollama:smoke -- <model>

Testing

• npm run build
• npm test
• npm run ollama:smoke -- llama3.2:3b

[patmakesapps]

Pushed a follow-up commit to address the CI failure around the
streamed Ollama proxy test.

What changed:

• switched the fetch proxy stream handling to ReadableStream.tee()
  so the caller gets a native response body while usage tracking reads a
  separate branch in the background
• added a bounded timeout in the NDJSON streaming test so CI fails
  explicitly instead of hanging on response.text()

Why:

• the failure looked like the wrapped response stream was not closing
  reliably on GitHub Actions, so response.text() could appear to get
  no response even though the fixture data was valid

Checks run locally:

• npm run build --workspace packages/core passed
• direct built-JS reproduction of the streamed Ollama path completed
  and tracked usage correctly
• full npm test could not be run in this sandbox because the local
  environment blocks child-process spawn (spawn EPERM)

Commit: 908ebc8