Make LlamaStackLibraryClient work correctly (#581)

This PR does a few things:

- it moves "direct client" to llama-stack repo instead of being in the
llama-stack-client-python repo
- renames it to `LlamaStackLibraryClient`
- actually makes synchronous generators work 
- makes streaming and non-streaming work properly

In many ways, this PR makes things finally "work"

## Test Plan

See a `library_client_test.py` I added. This isn't really quite a test
yet but it demonstrates that this mode now works. Here's the invocation
and the response:

```
INFERENCE_MODEL=meta-llama/Llama-3.2-3B-Instruct python llama_stack/distribution/tests/library_client_test.py ollama
```


![image](https://github.com/user-attachments/assets/17d4e116-4457-4755-a14e-d9a668801fe0)

llama_stack/providers/remote/inference/ollama/ollama.py

@@ -269,7 +269,6 @@ class OllamaInferenceAdapter(Inference, ModelsProtocolPrivate):
             r = await self.client.chat(**params)
         else:
             r = await self.client.generate(**params)
-        assert isinstance(r, dict)
 
         if "message" in r:
             choice = OpenAICompatCompletionChoice(