fix: Use LLAMA_STACK_CLIENT_TIMEOUT for OpenAI client in integration tests

The openai_client fixture had a hardcoded 30-second timeout, while the
integration test runner sets LLAMA_STACK_CLIENT_TIMEOUT=300. The
LlamaStackClient respected this env var but the OpenAI client did not,
causing timeouts when Ollama takes longer than 30 seconds to respond
on GitHub Actions runners.

Signed-off-by: Dan Prince <dprince@redhat.com>

Author: dprince

tests/integration/fixtures/common.py

@@ -376,7 +376,12 @@ def require_server(llama_stack_client):
 @pytest.fixture(scope="session")
 def openai_client(llama_stack_client, require_server):
     base_url = f"{llama_stack_client.base_url}/v1"
-    client = OpenAI(base_url=base_url, api_key="fake", max_retries=0, timeout=30.0)
+    client = OpenAI(
+        base_url=base_url,
+        api_key="fake",
+        max_retries=0,
+        timeout=float(os.environ.get("LLAMA_STACK_CLIENT_TIMEOUT", "60")),
+    )
     yield client
     # Cleanup: close HTTP connections
     try: