fix: make mlflow_logger import optional with no-op fallback

The strixhalo LLM worker uses py_executable pointing to the Docker
image venv which doesn't have the updated ray-serve-apps package.
Wrap all InferenceLogger imports in try/except and guard usage with
None checks so apps degrade gracefully without MLflow logging.

ray_serve/serve_tts.py

@@ -11,7 +11,10 @@ from typing import Any
 
 from ray import serve
 
-from ray_serve.mlflow_logger import InferenceLogger
+try:
+    from ray_serve.mlflow_logger import InferenceLogger
+except ImportError:
+    InferenceLogger = None
 
 
 @serve.deployment(name="TTSDeployment", num_replicas=1)
@@ -36,13 +39,16 @@ class TTSDeployment:
         print("TTS model loaded successfully")
 
         # MLflow metrics
-        self._mlflow = InferenceLogger(
-            experiment_name="ray-serve-tts",
-            run_name=f"tts-{self.model_name.split('/')[-1]}",
-            tags={"model.name": self.model_name, "model.framework": "coqui-tts", "gpu": str(self.use_gpu)},
-            flush_every=5,
-        )
-        self._mlflow.initialize(params={"model_name": self.model_name, "use_gpu": str(self.use_gpu)})
+        if InferenceLogger is not None:
+            self._mlflow = InferenceLogger(
+                experiment_name="ray-serve-tts",
+                run_name=f"tts-{self.model_name.split('/')[-1]}",
+                tags={"model.name": self.model_name, "model.framework": "coqui-tts", "gpu": str(self.use_gpu)},
+                flush_every=5,
+            )
+            self._mlflow.initialize(params={"model_name": self.model_name, "use_gpu": str(self.use_gpu)})
+        else:
+            self._mlflow = None
 
     async def __call__(self, request: dict[str, Any]) -> dict[str, Any]:
         """
@@ -104,12 +110,13 @@ class TTSDeployment:
             duration = len(wav) / sample_rate
 
             # Log to MLflow
-            self._mlflow.log_request(
-                latency_s=time.time() - _start,
-                audio_duration_s=duration,
-                text_chars=len(text),
-                realtime_factor=(time.time() - _start) / duration if duration > 0 else 0,
-            )
+            if self._mlflow:
+                self._mlflow.log_request(
+                    latency_s=time.time() - _start,
+                    audio_duration_s=duration,
+                    text_chars=len(text),
+                    realtime_factor=(time.time() - _start) / duration if duration > 0 else 0,
+                )
 
             response = {
                 "model": self.model_name,