Fix Stella models CPU compatibility by disabling xformers

Stella models were failing on CPU with xformers attention operator errors.
Added environment variable XFORMERS_DISABLED=1 for Stella models on CPU
to force fallback to standard PyTorch attention mechanisms.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

app.py

@@ -71,6 +71,13 @@ def load_model(model_name: str):
     trust_remote_code = MODELS.get(model_name, {}).get("trust_remote_code", False)
     try:
         print(f"Loading model '{model_name}' on {DEVICE}")
+        
+        # Special handling for Stella models on CPU
+        if "stella" in model_name.lower() and DEVICE == "cpu":
+            # Disable xformers for CPU inference to avoid attention operator errors
+            import os
+            os.environ["XFORMERS_DISABLED"] = "1"
+        
         current_model = SentenceTransformer(
             model_name, 
             trust_remote_code=trust_remote_code,