Resume turn after calling client tool (#102)

# What does this PR do?

When client tools are used, `create_turn` currently terminates after
executing the client tool requested by the model.
This PR proposes to mirror the behavior when only server side tools are
used where the tool response is passed back to the model for a final
response.


## Test Plan

Before:
<img width="1352" alt="image"
src="https://github.com/user-attachments/assets/90dc6bf4-9fa0-48e0-9e08-c31d3cc8ca44"
/>

After:
<img width="1372" alt="image"
src="https://github.com/user-attachments/assets/08e6bf64-1202-4ea8-a293-25b331488949"
/>


## Sources

Please link relevant resources if necessary.


## Before submitting

- [ ] This PR fixes a typo or improves the docs (you can dismiss the
other checks if that's the case).
- [ ] Ran pre-commit to handle lint / formatting issues.
- [ ] Read the [contributor
guideline](https://github.com/meta-llama/llama-stack/blob/main/CONTRIBUTING.md),
      Pull Request section?
- [ ] Updated relevant documentation.
- [ ] Wrote necessary unit or integration tests.

---------

Co-authored-by: Eric Huang (AI Platform) <erichuang@fb.com>

Author: ehhuang

src/llama_stack_client/lib/agents/agent.py

@@ -13,6 +13,7 @@
 
 from .client_tool import ClientTool
 
+DEFAULT_MAX_ITER = 10
 
 class Agent:
     def __init__(
@@ -75,21 +76,32 @@ def create_turn(
         toolgroups: Optional[List[Toolgroup]] = None,
         documents: Optional[List[Document]] = None,
     ):
-        response = self.client.agents.turn.create(
-            agent_id=self.agent_id,
-            # use specified session_id or last session created
-            session_id=session_id or self.session_id[-1],
-            messages=messages,
-            stream=True,
-            documents=documents,
-            toolgroups=toolgroups,
-        )
-        for chunk in response:
-            if hasattr(chunk, "error"):
-                yield chunk
-                return
-            elif not self._has_tool_call(chunk):
-                yield chunk
-            else:
-                next_message = self._run_tool(chunk)
-                yield next_message
+        stop = False
+        n_iter = 0
+        max_iter = self.agent_config.get('max_infer_iters', DEFAULT_MAX_ITER)
+        while not stop and n_iter < max_iter:
+            response = self.client.agents.turn.create(
+                agent_id=self.agent_id,
+                # use specified session_id or last session created
+                session_id=session_id or self.session_id[-1],
+                messages=messages,
+                stream=True,
+                documents=documents,
+                toolgroups=toolgroups,
+            )
+            # by default, we stop after the first turn
+            stop = True
+            for chunk in response:
+                if hasattr(chunk, "error"):
+                    yield chunk
+                    return
+                elif not self._has_tool_call(chunk):
+                    yield chunk
+                else:
+                    next_message = self._run_tool(chunk)
+                    yield next_message
+
+                    # continue the turn when there's a tool call
+                    stop = False
+                    messages = [next_message]
+                    n_iter += 1