Implemented optional concurrency for the LLM backend

tests/test_framework_llm.py

@@ -1,3 +1,4 @@
+from concurrent.futures import ThreadPoolExecutor
 import sys
 import types
 
@@ -16,6 +17,7 @@ def _config(**overrides):
     base = AuditaConfig.from_sources(env={})
     data = {
         "api_key": "test-key",
+        "llm_concurrency": base.llm_concurrency,
         "module_keys": base.module_keys,
         "model": base.model,
         "base_url": base.base_url,
@@ -172,3 +174,24 @@ def test_missing_api_key_error_is_provider_neutral():
             response_model=DummyResponseModel,
             config=_config(api_key=None),
         )
+
+
+def test_client_initialization_is_safe_under_concurrent_calls(monkeypatch):
+    _, openai_inits = _install_fake_llm_modules(monkeypatch)
+    client = OpenAICompatibleStructuredLLMClient()
+    config = _config(api_key="key-1", base_url="http://localhost:8000/v1")
+
+    with ThreadPoolExecutor(max_workers=4) as executor:
+        list(
+            executor.map(
+                lambda _: client.run_structured(
+                    stage_name="test-stage",
+                    messages=[{"role": "user", "content": "Hello"}],
+                    response_model=DummyResponseModel,
+                    config=config,
+                ),
+                range(4),
+            )
+        )
+
+    assert openai_inits == [{"api_key": "key-1", "base_url": "http://localhost:8000/v1"}]