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_embeddings.py

@@ -9,7 +9,10 @@ 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="EmbeddingsDeployment", num_replicas=1)
@@ -37,15 +40,18 @@ class EmbeddingsDeployment:
         print(f"Model loaded. Embedding dimension: {self.embedding_dim}")
 
         # MLflow metrics
-        self._mlflow = InferenceLogger(
-            experiment_name="ray-serve-embeddings",
-            run_name=f"embeddings-{self.model_id.split('/')[-1]}",
-            tags={"model.name": self.model_id, "model.framework": "sentence-transformers", "device": self.device},
-            flush_every=10,
-        )
-        self._mlflow.initialize(
-            params={"model_id": self.model_id, "embedding_dim": str(self.embedding_dim), "device": self.device}
-        )
+        if InferenceLogger is not None:
+            self._mlflow = InferenceLogger(
+                experiment_name="ray-serve-embeddings",
+                run_name=f"embeddings-{self.model_id.split('/')[-1]}",
+                tags={"model.name": self.model_id, "model.framework": "sentence-transformers", "device": self.device},
+                flush_every=10,
+            )
+            self._mlflow.initialize(
+                params={"model_id": self.model_id, "embedding_dim": str(self.embedding_dim), "device": self.device}
+            )
+        else:
+            self._mlflow = None
 
     async def __call__(self, request: dict[str, Any]) -> dict[str, Any]:
         """
@@ -86,11 +92,12 @@ class EmbeddingsDeployment:
             total_tokens += len(text.split())
 
         # Log to MLflow
-        self._mlflow.log_request(
-            latency_s=time.time() - _start,
-            batch_size=len(texts),
-            total_tokens=total_tokens,
-        )
+        if self._mlflow:
+            self._mlflow.log_request(
+                latency_s=time.time() - _start,
+                batch_size=len(texts),
+                total_tokens=total_tokens,
+            )
 
         # Return OpenAI-compatible response
         return {